X-Git-Url: https://perl5.git.perl.org/perl5.git/blobdiff_plain/ab3bbdeb874c2a82798e2c9cc4b61acf5866b410..7f69552c33ff9bd1eb6665f732d0f22956ac2f30:/regexec.c diff --git a/regexec.c b/regexec.c index 3eee31e..17d0e01 100644 --- a/regexec.c +++ b/regexec.c @@ -77,7 +77,7 @@ #define RF_tainted 1 /* tainted information used? */ #define RF_warned 2 /* warned about big count? */ -#define RF_evaled 4 /* Did an EVAL with setting? */ + #define RF_utf8 8 /* Pattern contains multibyte chars? */ #define UTF ((PL_reg_flags & RF_utf8) != 0) @@ -109,7 +109,7 @@ ? (U8*)pos - off \ : NULL) -#define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)pos, off, (U8*)lim) : (U8*)(pos + off)) +#define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off)) #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim)) #define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \ @@ -150,6 +150,7 @@ } \ } STMT_END + static void restore_pos(pTHX_ void *arg); STATIC CHECKPOINT @@ -165,8 +166,9 @@ S_regcppush(pTHX_ I32 parenfloor) if (paren_elems_to_push < 0) Perl_croak(aTHX_ "panic: paren_elems_to_push < 0"); -#define REGCP_OTHER_ELEMS 6 +#define REGCP_OTHER_ELEMS 8 SSGROW(paren_elems_to_push + REGCP_OTHER_ELEMS); + for (p = PL_regsize; p > parenfloor; p--) { /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */ SSPUSHINT(PL_regendp[p]); @@ -181,6 +183,8 @@ S_regcppush(pTHX_ I32 parenfloor) )); } /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */ + SSPUSHPTR(PL_regstartp); + SSPUSHPTR(PL_regendp); SSPUSHINT(PL_regsize); SSPUSHINT(*PL_reglastparen); SSPUSHINT(*PL_reglastcloseparen); @@ -197,17 +201,16 @@ S_regcppush(pTHX_ I32 parenfloor) /* These are needed since we do not localize EVAL nodes: */ #define REGCP_SET(cp) \ DEBUG_STATE_r( \ - if (cp != PL_savestack_ix) \ PerlIO_printf(Perl_debug_log, \ - " Setting an EVAL scope, savestack=%"IVdf"\n", \ + " Setting an EVAL scope, savestack=%"IVdf"\n", \ (IV)PL_savestack_ix)); \ cp = PL_savestack_ix #define REGCP_UNWIND(cp) \ - DEBUG_EXECUTE_r( \ + DEBUG_STATE_r( \ if (cp != PL_savestack_ix) \ - PerlIO_printf(Perl_debug_log, \ - " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \ + PerlIO_printf(Perl_debug_log, \ + " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \ (IV)(cp), (IV)PL_savestack_ix)); \ regcpblow(cp) @@ -228,7 +231,10 @@ S_regcppop(pTHX_ const regexp *rex) *PL_reglastcloseparen = SSPOPINT; *PL_reglastparen = SSPOPINT; PL_regsize = SSPOPINT; + PL_regendp=(I32 *) SSPOPPTR; + PL_regstartp=(I32 *) SSPOPPTR; + /* Now restore the parentheses context. */ for (i -= (REGCP_OTHER_ELEMS - REGCP_FRAME_ELEMS); i > 0; i -= REGCP_PAREN_ELEMS) { @@ -357,7 +363,7 @@ Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos, register SV *check; char *strbeg; char *t; - const int do_utf8 = sv ? SvUTF8(sv) : 0; /* if no sv we have to assume bytes */ + const bool do_utf8 = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */ I32 ml_anch; register char *other_last = NULL; /* other substr checked before this */ char *check_at = NULL; /* check substr found at this pos */ @@ -375,7 +381,8 @@ Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos, } DEBUG_EXECUTE_r( debug_start_match(prog, do_utf8, strpos, strend, - "Guessing start of match for"); + sv ? "Guessing start of match in sv for" + : "Guessing start of match in string for"); ); /* CHR_DIST() would be more correct here but it makes things slow. */ @@ -384,6 +391,7 @@ Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos, "String too short... [re_intuit_start]\n")); goto fail; } + strbeg = (sv && SvPOK(sv)) ? strend - SvCUR(sv) : strpos; PL_regeol = strend; if (do_utf8) { @@ -395,9 +403,9 @@ Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos, to_byte_substr(prog); check = prog->check_substr; } - if (check == &PL_sv_undef) { + if (check == &PL_sv_undef) { DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, - "Non-utf string cannot match utf check string\n")); + "Non-utf8 string cannot match utf8 check string\n")); goto fail; } if (prog->reganch & ROPT_ANCH) { /* Match at beg-of-str or after \n */ @@ -420,6 +428,7 @@ Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos, I32 slen; s = HOP3c(strpos, prog->check_offset_min, strend); + if (SvTAIL(check)) { slen = SvCUR(check); /* >= 1 */ @@ -449,8 +458,8 @@ Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos, /* Match is anchored, but substr is not anchored wrt beg-of-str. */ s = strpos; start_shift = prog->check_offset_min; /* okay to underestimate on CC */ - end_shift = prog->minlen - start_shift - - CHR_SVLEN(check) + (SvTAIL(check) != 0); + end_shift = prog->check_end_shift; + if (!ml_anch) { const I32 end = prog->check_offset_max + CHR_SVLEN(check) - (SvTAIL(check) != 0); @@ -463,20 +472,37 @@ Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos, else { /* Can match at random position */ ml_anch = 0; s = strpos; - start_shift = prog->check_offset_min; /* okay to underestimate on CC */ - /* Should be nonnegative! */ - end_shift = prog->minlen - start_shift - - CHR_SVLEN(check) + (SvTAIL(check) != 0); + start_shift = prog->check_offset_min; /* okay to underestimate on CC */ + end_shift = prog->check_end_shift; + + /* end shift should be non negative here */ } #ifdef DEBUGGING /* 7/99: reports of failure (with the older version) */ if (end_shift < 0) - Perl_croak(aTHX_ "panic: end_shift"); + Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ", + (IV)end_shift, prog->precomp); #endif restart: /* Find a possible match in the region s..strend by looking for the "check" substring in the region corrected by start/end_shift. */ + + { + I32 srch_start_shift = start_shift; + I32 srch_end_shift = end_shift; + if (srch_start_shift < 0 && strbeg - s > srch_start_shift) { + srch_end_shift -= ((strbeg - s) - srch_start_shift); + srch_start_shift = strbeg - s; + } + DEBUG_OPTIMISE_MORE_r({ + PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n", + (IV)prog->check_offset_min, + (IV)srch_start_shift, + (IV)srch_end_shift, + (IV)prog->check_end_shift); + }); + if (flags & REXEC_SCREAM) { I32 p = -1; /* Internal iterator of scream. */ I32 * const pp = data ? data->scream_pos : &p; @@ -486,7 +512,7 @@ Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos, && (BmPREVIOUS(check) == SvCUR(check) - 1) && SvTAIL(check) )) s = screaminstr(sv, check, - start_shift + (s - strbeg), end_shift, pp, 0); + srch_start_shift + (s - strbeg), srch_end_shift, pp, 0); else goto fail_finish; /* we may be pointing at the wrong string */ @@ -495,15 +521,27 @@ Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos, if (data) *data->scream_olds = s; } - else if (prog->reganch & ROPT_CANY_SEEN) - s = fbm_instr((U8*)(s + start_shift), - (U8*)(strend - end_shift), - check, multiline ? FBMrf_MULTILINE : 0); - else - s = fbm_instr(HOP3(s, start_shift, strend), - HOP3(strend, -end_shift, strbeg), + else { + U8* start_point; + U8* end_point; + if (prog->reganch & ROPT_CANY_SEEN) { + start_point= (U8*)(s + srch_start_shift); + end_point= (U8*)(strend - srch_end_shift); + } else { + start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend); + end_point= HOP3(strend, -srch_end_shift, strbeg); + } + DEBUG_OPTIMISE_MORE_r({ + PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n", + (int)(end_point - start_point), + (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point), + start_point); + }); + + s = fbm_instr( start_point, end_point, check, multiline ? FBMrf_MULTILINE : 0); - + } + } /* Update the count-of-usability, remove useless subpatterns, unshift s. */ @@ -521,12 +559,20 @@ Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos, if (!s) goto fail_finish; - - check_at = s; - /* Finish the diagnostic message */ DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) ); + /* XXX dmq: first branch is for positive lookbehind... + Our check string is offset from the beginning of the pattern. + So we need to do any stclass tests offset forward from that + point. I think. :-( + */ + + + + check_at=s; + + /* Got a candidate. Check MBOL anchoring, and the *other* substr. Start with the other substr. XXXX no SCREAM optimization yet - and a very coarse implementation @@ -535,7 +581,9 @@ Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos, Probably it is right to do no SCREAM here... */ - if (do_utf8 ? (prog->float_utf8 && prog->anchored_utf8) : (prog->float_substr && prog->anchored_substr)) { + if (do_utf8 ? (prog->float_utf8 && prog->anchored_utf8) + : (prog->float_substr && prog->anchored_substr)) + { /* Take into account the "other" substring. */ /* XXXX May be hopelessly wrong for UTF... */ if (!other_last) @@ -562,12 +610,17 @@ Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos, last2 = last1 = HOP3c(strend, -prog->minlen, strbeg); if (last < last1) last1 = last; - /* XXXX It is not documented what units *_offsets are in. Assume bytes. */ + /* XXXX It is not documented what units *_offsets are in. + We assume bytes, but this is clearly wrong. + Meaning this code needs to be carefully reviewed for errors. + dmq. + */ + /* On end-of-str: see comment below. */ must = do_utf8 ? prog->anchored_utf8 : prog->anchored_substr; if (must == &PL_sv_undef) { s = (char*)NULL; - DEBUG_EXECUTE_r(must = prog->anchored_utf8); /* for debug */ + DEBUG_r(must = prog->anchored_utf8); /* for debug */ } else s = fbm_instr( @@ -631,7 +684,7 @@ Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos, and end-of-str is not later than strend we are OK. */ if (must == &PL_sv_undef) { s = (char*)NULL; - DEBUG_EXECUTE_r(must = prog->float_utf8); /* for debug message */ + DEBUG_r(must = prog->float_utf8); /* for debug message */ } else s = fbm_instr((unsigned char*)s, @@ -670,11 +723,26 @@ Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos, } } - t = s - prog->check_offset_max; + + t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos); + + DEBUG_OPTIMISE_MORE_r( + PerlIO_printf(Perl_debug_log, + "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n", + (IV)prog->check_offset_min, + (IV)prog->check_offset_max, + (IV)(s-strpos), + (IV)(t-strpos), + (IV)(t-s), + (IV)(strend-strpos) + ) + ); + if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */ && (!do_utf8 - || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*)strpos)) - && t > strpos))) { + || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos))) + && t > strpos))) + { /* Fixed substring is found far enough so that the match cannot start at strpos. */ try_at_offset: @@ -783,7 +851,9 @@ Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos, /* Last resort... */ /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */ - if (prog->regstclass && OP(prog->regstclass)!=TRIE) { + /* trie stclasses are too expensive to use here, we are better off to + leave it to regmatch itself */ + if (prog->regstclass && PL_regkind[OP(prog->regstclass)]!=TRIE) { /* minlen == 0 is possible if regstclass is \b or \B, and the fixed substr is ''$. Since minlen is already taken into account, s+1 is before strend; @@ -796,14 +866,17 @@ Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos, const int cl_l = (PL_regkind[OP(prog->regstclass)] == EXACT ? CHR_DIST(str+STR_LEN(prog->regstclass), str) : 1); - const char * endpos = (prog->anchored_substr || prog->anchored_utf8 || ml_anch) - ? HOP3c(s, (prog->minlen ? cl_l : 0), strend) - : (prog->float_substr || prog->float_utf8 - ? HOP3c(HOP3c(check_at, -start_shift, strbeg), - cl_l, strend) - : strend); - /*if (OP(prog->regstclass) == TRIE) - endpos++;*/ + char * endpos; + if (prog->anchored_substr || prog->anchored_utf8 || ml_anch) + endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend); + else if (prog->float_substr || prog->float_utf8) + endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend); + else + endpos= strend; + + DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %d s: %d endpos: %d\n", + (IV)start_shift, check_at - strbeg, s - strbeg, endpos - strbeg)); + t = s; s = find_byclass(prog, prog->regstclass, s, endpos, NULL); if (!s) { @@ -896,10 +969,7 @@ Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos, return NULL; } -/* We know what class REx starts with. Try to find this position... */ -/* if reginfo is NULL, its a dryrun */ -/* annoyingly all the vars in this routine have different names from their counterparts - in regmatch. /grrr */ + #define REXEC_TRIE_READ_CHAR(trie_type, trie, uc, uscan, len, uvc, charid, \ foldlen, foldbuf, uniflags) STMT_START { \ @@ -1039,6 +1109,14 @@ if ((!reginfo || regtry(reginfo, s))) \ } \ break +#define DUMP_EXEC_POS(li,s,doutf8) \ + dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8) + +/* We know what class REx starts with. Try to find this position... */ +/* if reginfo is NULL, its a dryrun */ +/* annoyingly all the vars in this routine have different names from their counterparts + in regmatch. /grrr */ + STATIC char * S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s, const char *strend, const regmatch_info *reginfo) @@ -1325,8 +1403,8 @@ S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s, !isDIGIT_LC_utf8((U8*)s), !isDIGIT_LC(*s) ); - case TRIE: - /*Perl_croak(aTHX_ "panic: unknown regstclass TRIE");*/ + case AHOCORASICKC: + case AHOCORASICK: { const enum { trie_plain, trie_utf8, trie_utf8_fold } trie_type = do_utf8 ? @@ -1344,11 +1422,13 @@ S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s, STRLEN maxlen = trie->maxlen; SV *sv_points; U8 **points; /* map of where we were in the input string - when reading a given string. For ASCII this + when reading a given char. For ASCII this is unnecessary overhead as the relationship is always 1:1, but for unicode, especially case folded unicode this is not true. */ U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ]; + U8 *bitmap=NULL; + GET_RE_DEBUG_FLAGS_DECL; @@ -1363,13 +1443,31 @@ S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s, SvPOK_on(sv_points); sv_2mortal(sv_points); points=(U8**)SvPV_nolen(sv_points ); - - if (trie->bitmap && trie_type != trie_utf8_fold) { - while (s <= last_start && !TRIE_BITMAP_TEST(trie,*s) ) { - s++; - } + if ( trie_type != trie_utf8_fold + && (trie->bitmap || OP(c)==AHOCORASICKC) ) + { + if (trie->bitmap) + bitmap=(U8*)trie->bitmap; + else + bitmap=(U8*)ANYOF_BITMAP(c); } - + /* this is the Aho-Corasick algorithm modified a touch + to include special handling for long "unknown char" + sequences. The basic idea being that we use AC as long + as we are dealing with a possible matching char, when + we encounter an unknown char (and we have not encountered + an accepting state) we scan forward until we find a legal + starting char. + AC matching is basically that of trie matching, except + that when we encounter a failing transition, we fall back + to the current states "fail state", and try the current char + again, a process we repeat until we reach the root state, + state 1, or a legal transition. If we fail on the root state + then we can either terminate if we have reached an accepting + state previously, or restart the entire process from the beginning + if we have not. + + */ while (s <= last_start) { const U32 uniflags = UTF8_ALLOW_DEFAULT; U8 *uc = (U8*)s; @@ -1381,39 +1479,68 @@ S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s, STRLEN foldlen = 0; U8 *uscan = (U8*)NULL; U8 *leftmost = NULL; - +#ifdef DEBUGGING + U32 accepted_word= 0; +#endif U32 pointpos = 0; while ( state && uc <= (U8*)strend ) { int failed=0; - if (aho->states[ state ].wordnum) { - U8 *lpos= points[ (pointpos - trie->wordlen[aho->states[ state ].wordnum-1] ) % maxlen ]; - if (!leftmost || lpos < leftmost) + U32 word = aho->states[ state ].wordnum; + + if( state==1 ) { + if ( bitmap ) { + DEBUG_TRIE_EXECUTE_r( + if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) { + dump_exec_pos( (char *)uc, c, strend, real_start, + (char *)uc, do_utf8 ); + PerlIO_printf( Perl_debug_log, + " Scanning for legal start char...\n"); + } + ); + while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) { + uc++; + } + s= (char *)uc; + } + if (uc >(U8*)last_start) break; + } + + if ( word ) { + U8 *lpos= points[ (pointpos - trie->wordlen[word-1] ) % maxlen ]; + if (!leftmost || lpos < leftmost) { + DEBUG_r(accepted_word=word); leftmost= lpos; + } if (base==0) break; + } points[pointpos++ % maxlen]= uc; REXEC_TRIE_READ_CHAR(trie_type, trie, uc, uscan, len, uvc, charid, foldlen, foldbuf, uniflags); - DEBUG_TRIE_EXECUTE_r( + DEBUG_TRIE_EXECUTE_r({ + dump_exec_pos( (char *)uc, c, strend, real_start, + s, do_utf8 ); PerlIO_printf(Perl_debug_log, - "Pos: %d Charid:%3x CV:%4"UVxf" ", - (int)((const char*)uc - real_start), charid, uvc) - ); - uc += len; + " Charid:%3u CP:%4"UVxf" ", + charid, uvc); + }); do { #ifdef DEBUGGING - U32 word = aho->states[ state ].wordnum; + word = aho->states[ state ].wordnum; #endif base = aho->states[ state ].trans.base; - DEBUG_TRIE_EXECUTE_r( + DEBUG_TRIE_EXECUTE_r({ + if (failed) + dump_exec_pos( (char *)uc, c, strend, real_start, + s, do_utf8 ); PerlIO_printf( Perl_debug_log, - "%sState: %4"UVxf", Base: 0x%-4"UVxf" uvc=%"UVxf" word=%"UVxf"\n", - failed ? "Fail transition to " : "", - state, base, uvc, word) - ); + "%sState: %4"UVxf", word=%"UVxf, + failed ? " Fail transition to " : "", + (UV)state, (UV)word); + }); if ( base ) { U32 tmp; if (charid && @@ -1425,53 +1552,60 @@ S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s, && (tmp=trie->trans[base + charid - 1 - trie->uniquecharcount ].next)) { + DEBUG_TRIE_EXECUTE_r( + PerlIO_printf( Perl_debug_log," - legal\n")); state = tmp; break; } else { - failed++; - if ( state == 1 ) - break; - else - state = aho->fail[state]; + DEBUG_TRIE_EXECUTE_r( + PerlIO_printf( Perl_debug_log," - fail\n")); + failed = 1; + state = aho->fail[state]; } } else { /* we must be accepting here */ - failed++; + DEBUG_TRIE_EXECUTE_r( + PerlIO_printf( Perl_debug_log," - accepting\n")); + failed = 1; break; } } while(state); + uc += len; if (failed) { if (leftmost) break; - else if (!charid && trie->bitmap && trie_type != trie_utf8_fold) { - while ( uc <= (U8*)last_start && !TRIE_BITMAP_TEST(trie,*uc) ) { - uc++; - } - } + if (!state) state = 1; } } if ( aho->states[ state ].wordnum ) { U8 *lpos = points[ (pointpos - trie->wordlen[aho->states[ state ].wordnum-1]) % maxlen ]; - if (!leftmost || lpos < leftmost) + if (!leftmost || lpos < leftmost) { + DEBUG_r(accepted_word=aho->states[ state ].wordnum); leftmost = lpos; + } } - DEBUG_TRIE_EXECUTE_r( - PerlIO_printf( Perl_debug_log, - "%sState: %4"UVxf", Base: 0x%-4"UVxf" uvc=%"UVxf"\n", - "All done: ", - state, base, uvc) - ); if (leftmost) { s = (char*)leftmost; + DEBUG_TRIE_EXECUTE_r({ + PerlIO_printf( + Perl_debug_log,"Matches word #%"UVxf" at position %d. Trying full pattern...\n", + (UV)accepted_word, s - real_start + ); + }); if (!reginfo || regtry(reginfo, s)) { FREETMPS; LEAVE; goto got_it; } s = HOPc(s,1); + DEBUG_TRIE_EXECUTE_r({ + PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n"); + }); } else { + DEBUG_TRIE_EXECUTE_r( + PerlIO_printf( Perl_debug_log,"No match.\n")); break; } } @@ -1510,7 +1644,7 @@ Perl_regexec_flags(pTHX_ register regexp *prog, char *stringarg, register char * I32 scream_pos = -1; /* Internal iterator of scream. */ char *scream_olds = NULL; SV* const oreplsv = GvSV(PL_replgv); - const bool do_utf8 = DO_UTF8(sv); + const bool do_utf8 = (bool)DO_UTF8(sv); I32 multiline; regmatch_info reginfo; /* create some info to pass to regtry etc */ @@ -1529,14 +1663,20 @@ Perl_regexec_flags(pTHX_ register regexp *prog, char *stringarg, register char * reginfo.prog = prog; RX_MATCH_UTF8_set(prog, do_utf8); + DEBUG_EXECUTE_r( + debug_start_match(prog, do_utf8, startpos, strend, + "Matching"); + ); minlen = prog->minlen; - if (strend - startpos < minlen) { + + if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) { DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too short [regexec_flags]...\n")); goto phooey; } + /* Check validity of program. */ if (UCHARAT(prog->program) != REG_MAGIC) { Perl_croak(aTHX_ "corrupted regexp program"); @@ -1595,10 +1735,7 @@ Perl_regexec_flags(pTHX_ register regexp *prog, char *stringarg, register char * } } - DEBUG_EXECUTE_r( - debug_start_match(prog, do_utf8, startpos, strend, - "Matching"); - ); + /* Simplest case: anchored match need be tried only once. */ /* [unless only anchor is BOL and multiline is set] */ @@ -1643,7 +1780,11 @@ Perl_regexec_flags(pTHX_ register regexp *prog, char *stringarg, register char * } } goto phooey; - } else if (prog->reganch & ROPT_ANCH_GPOS) { + } else if (ROPT_GPOS_CHECK == (prog->reganch & ROPT_GPOS_CHECK)) + { + /* the warning about reginfo.ganch being used without intialization + is bogus -- we set it above, when prog->reganch & ROPT_GPOS_SEEN + and we only enter this block when the same bit is set. */ if (regtry(®info, reginfo.ganch)) goto got_it; goto phooey; @@ -1712,14 +1853,19 @@ Perl_regexec_flags(pTHX_ register regexp *prog, char *stringarg, register char * back_max = prog->float_max_offset; back_min = prog->float_min_offset; } + + if (must == &PL_sv_undef) /* could not downgrade utf8 check substring, so must fail */ goto phooey; - last = HOP3c(strend, /* Cannot start after this */ - -(I32)(CHR_SVLEN(must) - - (SvTAIL(must) != 0) + back_min), strbeg); - + if (back_min<0) { + last = strend; + } else { + last = HOP3c(strend, /* Cannot start after this */ + -(I32)(CHR_SVLEN(must) + - (SvTAIL(must) != 0) + back_min), strbeg); + } if (s > PL_bostr) last1 = HOPc(s, -1); else @@ -1732,9 +1878,9 @@ Perl_regexec_flags(pTHX_ register regexp *prog, char *stringarg, register char * strend = HOPc(strend, -dontbother); while ( (s <= last) && ((flags & REXEC_SCREAM) - ? (s = screaminstr(sv, must, HOP3c(s, back_min, strend) - strbeg, + ? (s = screaminstr(sv, must, HOP3c(s, back_min, (back_min<0 ? strbeg : strend)) - strbeg, end_shift, &scream_pos, 0)) - : (s = fbm_instr((unsigned char*)HOP3(s, back_min, strend), + : (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)), (unsigned char*)strend, must, multiline ? FBMrf_MULTILINE : 0))) ) { /* we may be pointing at the wrong string */ @@ -1776,11 +1922,11 @@ Perl_regexec_flags(pTHX_ register regexp *prog, char *stringarg, register char * }); goto phooey; } - else if ((c = prog->regstclass)) { + else if ( (c = prog->regstclass) ) { if (minlen) { const OPCODE op = OP(prog->regstclass); /* don't bother with what can't match */ - if (PL_regkind[op] != EXACT && op != CANY && op != TRIE) + if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE) strend = HOPc(strend, -(minlen - 1)); } DEBUG_EXECUTE_r({ @@ -1791,7 +1937,7 @@ Perl_regexec_flags(pTHX_ register regexp *prog, char *stringarg, register char * s,strend-s,60); PerlIO_printf(Perl_debug_log, "Matching stclass %.*s against %s (%d chars)\n", - SvCUR(prop), SvPVX_const(prop), + (int)SvCUR(prop), SvPVX_const(prop), quoted, (int)(strend - s)); } }); @@ -1840,9 +1986,10 @@ Perl_regexec_flags(pTHX_ register regexp *prog, char *stringarg, register char * } } if (last == NULL) { - DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, - "%sCan't trim the tail, match fails (should not happen)%s\n", - PL_colors[4], PL_colors[5])); + DEBUG_EXECUTE_r( + PerlIO_printf(Perl_debug_log, + "%sCan't trim the tail, match fails (should not happen)%s\n", + PL_colors[4], PL_colors[5])); goto phooey; /* Should not happen! */ } dontbother = strend - last + prog->float_min_offset; @@ -1882,6 +2029,8 @@ got_it: the same. */ restore_pos(aTHX_ prog); } + if (prog->paren_names) + (void)hv_iterinit(prog->paren_names); /* make sure $`, $&, $', and $digit will work later */ if ( !(flags & REXEC_NOT_FIRST) ) { @@ -1924,6 +2073,7 @@ phooey: return 0; } + /* - regtry - try match at specific point */ @@ -1937,9 +2087,6 @@ S_regtry(pTHX_ const regmatch_info *reginfo, char *startpos) regexp *prog = reginfo->prog; GET_RE_DEBUG_FLAGS_DECL; -#ifdef DEBUGGING - PL_regindent = 0; /* XXXX Not good when matches are reenterable... */ -#endif if ((prog->reganch & ROPT_EVAL_SEEN) && !PL_reg_eval_set) { MAGIC *mg; @@ -1948,7 +2095,7 @@ S_regtry(pTHX_ const regmatch_info *reginfo, char *startpos) PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n", (IV)(PL_stack_sp - PL_stack_base)); )); - SAVEI32(cxstack[cxstack_ix].blk_oldsp); + SAVESTACK_CXPOS(); cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base; /* Otherwise OP_NEXTSTATE will free whatever on stack now. */ SAVETMPS; @@ -2010,16 +2157,16 @@ S_regtry(pTHX_ const regmatch_info *reginfo, char *startpos) prog->subbeg = PL_bostr; prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */ } + DEBUG_EXECUTE_r(PL_reg_starttry = startpos); prog->startp[0] = startpos - PL_bostr; PL_reginput = startpos; - PL_regstartp = prog->startp; - PL_regendp = prog->endp; PL_reglastparen = &prog->lastparen; PL_reglastcloseparen = &prog->lastcloseparen; prog->lastparen = 0; prog->lastcloseparen = 0; PL_regsize = 0; - DEBUG_EXECUTE_r(PL_reg_starttry = startpos); + PL_regstartp = prog->startp; + PL_regendp = prog->endp; if (PL_reg_start_tmpl <= prog->nparens) { PL_reg_start_tmpl = prog->nparens*3/2 + 3; if(PL_reg_start_tmp) @@ -2065,27 +2212,20 @@ S_regtry(pTHX_ const regmatch_info *reginfo, char *startpos) #define sayYES goto yes #define sayNO goto no -#define sayNO_ANYOF goto no_anyof -#define sayYES_FINAL goto yes_final -#define sayNO_FINAL goto no_final -#define sayNO_SILENT goto do_no -#define saySAME(x) if (x) goto yes; else goto no - -#define CACHEsayNO STMT_START { \ - if (st->u.whilem.cache_offset | st->u.whilem.cache_bit) \ - PL_reg_poscache[st->u.whilem.cache_offset] |= \ - (1<u.whilem.cache_bit); \ - sayNO; \ -} STMT_END +#define sayNO_SILENT goto no_silent +/* we dont use STMT_START/END here because it leads to + "unreachable code" warnings, which are bogus, but distracting. */ +#define CACHEsayNO \ + if (ST.cache_mask) \ + PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \ + sayNO /* this is used to determine how far from the left messages like - 'failed...' are printed. Currently 29 makes these messages line - up with the opcode they refer to. Earlier perls used 25 which - left these messages outdented making reviewing a debug output - quite difficult. + 'failed...' are printed. It should be set such that messages + are inline with the regop output that created them. */ -#define REPORT_CODE_OFF 29 +#define REPORT_CODE_OFF 32 /* Make sure there is a test for this +1 options in re_tests */ @@ -2116,14 +2256,6 @@ S_push_slab(pTHX) return SLAB_FIRST(s); } -/* simulate a recursive call to regmatch */ - -#define REGMATCH(ns, where) \ - st->scan = scan; \ - scan = (ns); \ - st->resume_state = resume_##where; \ - goto start_recurse; \ - resume_point_##where: /* push a new state then goto it */ @@ -2142,141 +2274,153 @@ S_push_slab(pTHX) /* - - regmatch - main matching routine - * - * Conceptually the strategy is simple: check to see whether the current - * node matches, call self recursively to see whether the rest matches, - * and then act accordingly. In practice we make some effort to avoid - * recursion, in particular by going through "ordinary" nodes (that don't - * need to know whether the rest of the match failed) by a loop instead of - * by recursion. - */ -/* [lwall] I've hoisted the register declarations to the outer block in order to - * maybe save a little bit of pushing and popping on the stack. It also takes - * advantage of machines that use a register save mask on subroutine entry. - * - * This function used to be heavily recursive, but since this had the - * effect of blowing the CPU stack on complex regexes, it has been - * restructured to be iterative, and to save state onto the heap rather - * than the stack. Essentially whereever regmatch() used to be called, it - * pushes the current state, notes where to return, then jumps back into - * the main loop. - * - * Originally the structure of this function used to look something like - S_regmatch() { - int a = 1, b = 2; +regmatch() - main matching routine + +This is basically one big switch statement in a loop. We execute an op, +set 'next' to point the next op, and continue. If we come to a point which +we may need to backtrack to on failure such as (A|B|C), we push a +backtrack state onto the backtrack stack. On failure, we pop the top +state, and re-enter the loop at the state indicated. If there are no more +states to pop, we return failure. + +Sometimes we also need to backtrack on success; for example /A+/, where +after successfully matching one A, we need to go back and try to +match another one; similarly for lookahead assertions: if the assertion +completes successfully, we backtrack to the state just before the assertion +and then carry on. In these cases, the pushed state is marked as +'backtrack on success too'. This marking is in fact done by a chain of +pointers, each pointing to the previous 'yes' state. On success, we pop to +the nearest yes state, discarding any intermediate failure-only states. +Sometimes a yes state is pushed just to force some cleanup code to be +called at the end of a successful match or submatch; e.g. (??{$re}) uses +it to free the inner regex. + +Note that failure backtracking rewinds the cursor position, while +success backtracking leaves it alone. + +A pattern is complete when the END op is executed, while a subpattern +such as (?=foo) is complete when the SUCCESS op is executed. Both of these +ops trigger the "pop to last yes state if any, otherwise return true" +behaviour. + +A common convention in this function is to use A and B to refer to the two +subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is +the subpattern to be matched possibly multiple times, while B is the entire +rest of the pattern. Variable and state names reflect this convention. + +The states in the main switch are the union of ops and failure/success of +substates associated with with that op. For example, IFMATCH is the op +that does lookahead assertions /(?=A)B/ and so the IFMATCH state means +'execute IFMATCH'; while IFMATCH_A is a state saying that we have just +successfully matched A and IFMATCH_A_fail is a state saying that we have +just failed to match A. Resume states always come in pairs. The backtrack +state we push is marked as 'IFMATCH_A', but when that is popped, we resume +at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking +on success or failure. + +The struct that holds a backtracking state is actually a big union, with +one variant for each major type of op. The variable st points to the +top-most backtrack struct. To make the code clearer, within each +block of code we #define ST to alias the relevant union. + +Here's a concrete example of a (vastly oversimplified) IFMATCH +implementation: + + switch (state) { + .... + +#define ST st->u.ifmatch + + case IFMATCH: // we are executing the IFMATCH op, (?=A)B + ST.foo = ...; // some state we wish to save ... - while (scan != NULL) { - a++; // do stuff with a and b - ... - switch (OP(scan)) { - case FOO: { - int local = 3; - ... - if (regmatch(...)) // recurse - goto yes; - } - ... - } - } - yes: - return 1; - } + // push a yes backtrack state with a resume value of + // IFMATCH_A/IFMATCH_A_fail, then continue execution at the + // first node of A: + PUSH_YES_STATE_GOTO(IFMATCH_A, A); + // NOTREACHED + + case IFMATCH_A: // we have successfully executed A; now continue with B + next = B; + bar = ST.foo; // do something with the preserved value + break; + + case IFMATCH_A_fail: // A failed, so the assertion failed + ...; // do some housekeeping, then ... + sayNO; // propagate the failure - * Now it looks something like this: +#undef ST + + ... + } - typedef struct { - int a, b, local; - int resume_state; - } regmatch_state; +For any old-timers reading this who are familiar with the old recursive +approach, the code above is equivalent to: - S_regmatch() { - regmatch_state *st = new(); - int depth=0; - st->a++; // do stuff with a and b + case IFMATCH: // we are executing the IFMATCH op, (?=A)B + { + int foo = ... ... - while (scan != NULL) { - ... - switch (OP(scan)) { - case FOO: { - st->local = 3; - ... - st->scan = scan; - scan = ...; - st->resume_state = resume_FOO; - goto start_recurse; // recurse - - resume_point_FOO: - if (result) - goto yes; - } - ... - } - start_recurse: - st = new(); push a new state - st->a = 1; st->b = 2; - depth++; - } - yes: - result = 1; - if (depth--) { - st = pop(); - switch (resume_state) { - case resume_FOO: - goto resume_point_FOO; - ... - } + if (regmatch(A)) { + next = B; + bar = foo; + break; } - return result + ...; // do some housekeeping, then ... + sayNO; // propagate the failure } - - * WARNING: this means that any line in this function that contains a - * REGMATCH() or TRYPAREN() is actually simulating a recursive call to - * regmatch() using gotos instead. Thus the values of any local variables - * not saved in the regmatch_state structure will have been lost when - * execution resumes on the next line . - * - * States (ie the st pointer) are allocated in slabs of about 4K in size. - * PL_regmatch_state always points to the currently active state, and - * PL_regmatch_slab points to the slab currently containing PL_regmatch_state. - * The first time regmatch is called, the first slab is allocated, and is - * never freed until interpreter desctruction. When the slab is full, - * a new one is allocated chained to the end. At exit from regmatch, slabs - * allocated since entry are freed. - */ + +The topmost backtrack state, pointed to by st, is usually free. If you +want to claim it, populate any ST.foo fields in it with values you wish to +save, then do one of + + PUSH_STATE_GOTO(resume_state, node); + PUSH_YES_STATE_GOTO(resume_state, node); + +which sets that backtrack state's resume value to 'resume_state', pushes a +new free entry to the top of the backtrack stack, then goes to 'node'. +On backtracking, the free slot is popped, and the saved state becomes the +new free state. An ST.foo field in this new top state can be temporarily +accessed to retrieve values, but once the main loop is re-entered, it +becomes available for reuse. + +Note that the depth of the backtrack stack constantly increases during the +left-to-right execution of the pattern, rather than going up and down with +the pattern nesting. For example the stack is at its maximum at Z at the +end of the pattern, rather than at X in the following: + + /(((X)+)+)+....(Y)+....Z/ + +The only exceptions to this are lookahead/behind assertions and the cut, +(?>A), which pop all the backtrack states associated with A before +continuing. -/* *** every FOO_fail should = FOO+1 */ -#define TRIE_next (REGNODE_MAX+1) -#define TRIE_next_fail (REGNODE_MAX+2) -#define EVAL_A (REGNODE_MAX+3) -#define EVAL_A_fail (REGNODE_MAX+4) -#define resume_CURLYX (REGNODE_MAX+5) -#define resume_WHILEM1 (REGNODE_MAX+6) -#define resume_WHILEM2 (REGNODE_MAX+7) -#define resume_WHILEM3 (REGNODE_MAX+8) -#define resume_WHILEM4 (REGNODE_MAX+9) -#define resume_WHILEM5 (REGNODE_MAX+10) -#define resume_WHILEM6 (REGNODE_MAX+11) -#define BRANCH_next (REGNODE_MAX+12) -#define BRANCH_next_fail (REGNODE_MAX+13) -#define CURLYM_A (REGNODE_MAX+14) -#define CURLYM_A_fail (REGNODE_MAX+15) -#define CURLYM_B (REGNODE_MAX+16) -#define CURLYM_B_fail (REGNODE_MAX+17) -#define IFMATCH_A (REGNODE_MAX+18) -#define IFMATCH_A_fail (REGNODE_MAX+19) -#define CURLY_B_min_known (REGNODE_MAX+20) -#define CURLY_B_min_known_fail (REGNODE_MAX+21) -#define CURLY_B_min (REGNODE_MAX+22) -#define CURLY_B_min_fail (REGNODE_MAX+23) -#define CURLY_B_max (REGNODE_MAX+24) -#define CURLY_B_max_fail (REGNODE_MAX+25) +Bascktrack state structs are allocated in slabs of about 4K in size. +PL_regmatch_state and st always point to the currently active state, +and PL_regmatch_slab points to the slab currently containing +PL_regmatch_state. The first time regmatch() is called, the first slab is +allocated, and is never freed until interpreter destruction. When the slab +is full, a new one is allocated and chained to the end. At exit from +regmatch(), slabs allocated since entry are freed. + +*/ + + +#define DEBUG_STATE_pp(pp) \ + DEBUG_STATE_r({ \ + DUMP_EXEC_POS(locinput, scan, do_utf8); \ + PerlIO_printf(Perl_debug_log, \ + " %*s"pp" %s\n", \ + depth*2, "", \ + reg_name[st->resume_state] ); \ + }); #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1) #ifdef DEBUGGING + STATIC void S_debug_start_match(pTHX_ const regexp *prog, const bool do_utf8, const char *start, const char *end, const char *blurb) @@ -2296,35 +2440,41 @@ S_debug_start_match(pTHX_ const regexp *prog, const bool do_utf8, PL_colors[4], blurb, PL_colors[5], s0, s1); if (do_utf8||utf8_pat) - PerlIO_printf(Perl_debug_log, "UTF-8 %s...\n", - !do_utf8 ? "pattern" : !utf8_pat ? "string" : - "pattern and string" + PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n", + utf8_pat ? "pattern" : "", + utf8_pat && do_utf8 ? " and " : "", + do_utf8 ? "string" : "" ); } } STATIC void -S_dump_exec_pos(pTHX_ const char *locinput, const regnode *scan, const bool do_utf8) +S_dump_exec_pos(pTHX_ const char *locinput, + const regnode *scan, + const char *loc_regeol, + const char *loc_bostr, + const char *loc_reg_starttry, + const bool do_utf8) { - const int docolor = *PL_colors[0]; + const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4]; const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */ - int l = (PL_regeol - locinput) > taill ? taill : (PL_regeol - locinput); + int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput); /* The part of the string before starttry has one color (pref0_len chars), between starttry and current position another one (pref_len - pref0_len chars), after the current position the third one. We assume that pref0_len <= pref_len, otherwise we decrease pref0_len. */ - int pref_len = (locinput - PL_bostr) > (5 + taill) - l - ? (5 + taill) - l : locinput - PL_bostr; + int pref_len = (locinput - loc_bostr) > (5 + taill) - l + ? (5 + taill) - l : locinput - loc_bostr; int pref0_len; while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len))) pref_len++; - pref0_len = pref_len - (locinput - PL_reg_starttry); - if (l + pref_len < (5 + taill) && l < PL_regeol - locinput) - l = ( PL_regeol - locinput > (5 + taill) - pref_len - ? (5 + taill) - pref_len : PL_regeol - locinput); + pref0_len = pref_len - (locinput - loc_reg_starttry); + if (l + pref_len < (5 + taill) && l < loc_regeol - locinput) + l = ( loc_regeol - locinput > (5 + taill) - pref_len + ? (5 + taill) - pref_len : loc_regeol - locinput); while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput + l))) l--; if (pref0_len < 0) @@ -2342,22 +2492,47 @@ S_dump_exec_pos(pTHX_ const char *locinput, const regnode *scan, const bool do_u pref_len - pref0_len, 60, 2, 3); RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2), - locinput, PL_regeol - locinput, 60, 0, 1); + locinput, loc_regeol - locinput, 10, 0, 1); + const STRLEN tlen=len0+len1+len2; PerlIO_printf(Perl_debug_log, "%4"IVdf" <%.*s%.*s%s%.*s>%*s|", - (IV)(locinput - PL_bostr), + (IV)(locinput - loc_bostr), len0, s0, len1, s1, (docolor ? "" : "> <"), len2, s2, - 15 - l - pref_len + 1, + (int)(tlen > 19 ? 0 : 19 - tlen), ""); } } #endif +/* reg_check_named_buff_matched() + * Checks to see if a named buffer has matched. The data array of + * buffer numbers corresponding to the buffer is expected to reside + * in the regexp->data->data array in the slot stored in the ARG() of + * node involved. Note that this routine doesn't actually care about the + * name, that information is not preserved from compilation to execution. + * Returns the index of the leftmost defined buffer with the given name + * or 0 if non of the buffers matched. + */ +STATIC I32 +S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan) { + I32 n; + SV *sv_dat=(SV*)rex->data->data[ ARG( scan ) ]; + I32 *nums=(I32*)SvPVX(sv_dat); + for ( n=0; n= nums[n] && + PL_regendp[nums[n]] != -1) + { + return nums[n]; + } + } + return 0; +} + STATIC I32 /* 0 failure, 1 success */ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) { @@ -2379,20 +2554,37 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) /* cache heavy used fields of st in registers */ register regnode *scan; register regnode *next; - register I32 n = 0; /* initialize to shut up compiler warning */ + register I32 n = 0; /* general value; init to avoid compiler warning */ + register I32 ln = 0; /* len or last; init to avoid compiler warning */ register char *locinput = PL_reginput; - - /* these variables are NOT saved during a recusive RFEGMATCH: */ register I32 nextchr; /* is always set to UCHARAT(locinput) */ - bool result; /* return value of S_regmatch */ - int depth = 0; /* depth of recursion */ + + bool result = 0; /* return value of S_regmatch */ + int depth = 0; /* depth of backtrack stack */ + int nochange_depth = 0; /* depth of RECURSE recursion with nochange*/ regmatch_state *yes_state = NULL; /* state to pop to on success of subpattern */ + regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */ + struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */ U32 state_num; - + + /* these three flags are set by various ops to signal information to + * the very next op. They have a useful lifetime of exactly one loop + * iteration, and are not preserved or restored by state pushes/pops + */ + bool sw = 0; /* the condition value in (?(cond)a|b) */ + bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */ + int logical = 0; /* the following EVAL is: + 0: (?{...}) + 1: (?(?{...})X|Y) + 2: (??{...}) + or the following IFMATCH/UNLESSM is: + false: plain (?=foo) + true: used as a condition: (?(?=foo)) + */ + #ifdef DEBUGGING GET_RE_DEBUG_FLAGS_DECL; - PL_regindent++; #endif /* on first ever call to regmatch, allocate first slab */ @@ -2413,10 +2605,6 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) if (st > SLAB_LAST(PL_regmatch_slab)) st = PL_regmatch_state = S_push_slab(aTHX); - st->minmod = 0; - st->sw = 0; - st->logical = 0; - st->cc = NULL; /* Note that nextchr is a byte even in UTF */ nextchr = UCHARAT(locinput); scan = prog; @@ -2424,14 +2612,16 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) DEBUG_EXECUTE_r( { SV * const prop = sv_newmortal(); - dump_exec_pos( locinput, scan, do_utf8 ); + regnode *rnext=regnext(scan); + DUMP_EXEC_POS( locinput, scan, do_utf8 ); regprop(rex, prop, scan); PerlIO_printf(Perl_debug_log, "%3"IVdf":%*s%s(%"IVdf")\n", - (IV)(scan - rex->program), PL_regindent*2, "", + (IV)(scan - rex->program), depth*2, "", SvPVX_const(prop), - PL_regkind[OP(scan)] == END ? 0 : (IV)(regnext(scan) - rex->program)); + (PL_regkind[OP(scan)] == END || !rnext) ? + 0 : (IV)(rnext - rex->program)); }); next = scan + NEXT_OFF(scan); @@ -2512,7 +2702,22 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) #undef ST #define ST st->u.trie - + case TRIEC: + /* In this case the charclass data is available inline so + we can fail fast without a lot of extra overhead. + */ + if (scan->flags == EXACT || !do_utf8) { + if(!ANYOF_BITMAP_TEST(scan, *locinput)) { + DEBUG_EXECUTE_r( + PerlIO_printf(Perl_debug_log, + "%*s %sfailed to match trie start class...%s\n", + REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]) + ); + sayNO_SILENT; + /* NOTREACHED */ + } + } + /* FALL THROUGH */ case TRIE: { /* what type of TRIE am I? (utf8 makes this contextual) */ @@ -2526,23 +2731,6 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) = (reg_trie_data*)rex->data->data[ ARG( scan ) ]; U32 state = trie->startstate; - U8 *uc = ( U8* )locinput; - U16 charid = 0; - U32 base = 0; - UV uvc = 0; - STRLEN len = 0; - STRLEN foldlen = 0; - U8 *uscan = (U8*)NULL; - STRLEN bufflen=0; - SV *sv_accept_buff = NULL; - U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ]; - - ST.accepted = 0; /* how many accepting states we have seen */ - ST.B = next; -#ifdef DEBUGGING - ST.me = scan; -#endif - if (trie->bitmap && trie_type != trie_utf8_fold && !TRIE_BITMAP_TEST(trie,*locinput) ) { @@ -2550,42 +2738,66 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, "%*s %smatched empty string...%s\n", - REPORT_CODE_OFF+PL_regindent*2, "", PL_colors[4], PL_colors[5]) + REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]) ); break; } else { DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, - "%*s %sfailed to match start class...%s\n", - REPORT_CODE_OFF+PL_regindent*2, "", PL_colors[4], PL_colors[5]) + "%*s %sfailed to match trie start class...%s\n", + REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]) ); sayNO_SILENT; } } + { + U8 *uc = ( U8* )locinput; + + STRLEN len = 0; + STRLEN foldlen = 0; + U8 *uscan = (U8*)NULL; + STRLEN bufflen=0; + SV *sv_accept_buff = NULL; + U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ]; + + ST.accepted = 0; /* how many accepting states we have seen */ + ST.B = next; + ST.jump = trie->jump; + ST.me = scan; + /* traverse the TRIE keeping track of all accepting states we transition through until we get to a failing node. */ while ( state && uc <= (U8*)PL_regeol ) { + U32 base = trie->states[ state ].trans.base; + UV uvc = 0; + U16 charid; + /* We use charid to hold the wordnum as we don't use it + for charid until after we have done the wordnum logic. + We define an alias just so that the wordnum logic reads + more naturally. */ + +#define got_wordnum charid + got_wordnum = trie->states[ state ].wordnum; - if (trie->states[ state ].wordnum) { - if (!ST.accepted ) { + if ( got_wordnum ) { + if ( ! ST.accepted ) { ENTER; SAVETMPS; bufflen = TRIE_INITAL_ACCEPT_BUFFLEN; sv_accept_buff=newSV(bufflen * sizeof(reg_trie_accepted) - 1); - SvCUR_set(sv_accept_buff, - sizeof(reg_trie_accepted)); + SvCUR_set(sv_accept_buff, 0); SvPOK_on(sv_accept_buff); sv_2mortal(sv_accept_buff); SAVETMPS; ST.accept_buff = (reg_trie_accepted*)SvPV_nolen(sv_accept_buff ); } - else { + do { if (ST.accepted >= bufflen) { bufflen *= 2; ST.accept_buff =(reg_trie_accepted*) @@ -2594,20 +2806,21 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) } SvCUR_set(sv_accept_buff,SvCUR(sv_accept_buff) + sizeof(reg_trie_accepted)); - } - ST.accept_buff[ST.accepted].wordnum = trie->states[state].wordnum; - ST.accept_buff[ST.accepted].endpos = uc; - ++ST.accepted; - } - base = trie->states[ state ].trans.base; + + ST.accept_buff[ST.accepted].wordnum = got_wordnum; + ST.accept_buff[ST.accepted].endpos = uc; + ++ST.accepted; + } while (trie->nextword && (got_wordnum= trie->nextword[got_wordnum])); + } +#undef got_wordnum DEBUG_TRIE_EXECUTE_r({ - dump_exec_pos( (char *)uc, scan, do_utf8 ); + DUMP_EXEC_POS( (char *)uc, scan, do_utf8 ); PerlIO_printf( Perl_debug_log, - "%*s %sState: %4"UVxf", Base: %4"UVxf", Accepted: %4"UVxf" ", - 2+PL_regindent * 2, "", PL_colors[4], - (UV)state, (UV)base, (UV)ST.accepted ); + "%*s %sState: %4"UVxf" Accepted: %4"UVxf" ", + 2+depth * 2, "", PL_colors[4], + (UV)state, (UV)ST.accepted ); }); if ( base ) { @@ -2635,7 +2848,7 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) } DEBUG_TRIE_EXECUTE_r( PerlIO_printf( Perl_debug_log, - "Charid:%3x CV:%4"UVxf" After State: %4"UVxf"%s\n", + "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n", charid, uvc, (UV)state, PL_colors[5] ); ); } @@ -2645,10 +2858,10 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) DEBUG_EXECUTE_r( PerlIO_printf( Perl_debug_log, "%*s %sgot %"IVdf" possible matches%s\n", - REPORT_CODE_OFF + PL_regindent * 2, "", + REPORT_CODE_OFF + depth * 2, "", PL_colors[4], (IV)ST.accepted, PL_colors[5] ); ); - } + }} /* FALL THROUGH */ @@ -2664,7 +2877,7 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) : NULL; PerlIO_printf( Perl_debug_log, "%*s %sonly one match left: #%d <%s>%s\n", - REPORT_CODE_OFF+PL_regindent*2, "", PL_colors[4], + REPORT_CODE_OFF+depth*2, "", PL_colors[4], ST.accept_buff[ 0 ].wordnum, tmp ? SvPV_nolen_const( *tmp ) : "not compiled under -Dr", PL_colors[5] ); @@ -2676,7 +2889,12 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) LEAVE; locinput = PL_reginput; nextchr = UCHARAT(locinput); - scan = ST.B; + + if ( !ST.jump || !ST.jump[ST.accept_buff[0].wordnum]) + scan = ST.B; + else + scan = ST.me + ST.jump[ST.accept_buff[0].wordnum]; + continue; /* execute rest of RE */ } @@ -2705,7 +2923,7 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) DEBUG_TRIE_EXECUTE_r( PerlIO_printf( Perl_debug_log, "%*s %sgot %"IVdf" (%d) as best, looking at %"IVdf" (%d)%s\n", - REPORT_CODE_OFF + PL_regindent * 2, "", PL_colors[4], + REPORT_CODE_OFF + depth * 2, "", PL_colors[4], (IV)best, ST.accept_buff[ best ].wordnum, (IV)cur, ST.accept_buff[ cur ].wordnum, PL_colors[5] ); ); @@ -2721,10 +2939,15 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) SV ** const tmp = RX_DEBUG(reginfo->prog) ? av_fetch( trie->words, ST.accept_buff[ best ].wordnum - 1, 0 ) : NULL; - PerlIO_printf( Perl_debug_log, "%*s %strying alternation #%d <%s> at node #%d %s\n", - REPORT_CODE_OFF+PL_regindent*2, "", PL_colors[4], + regnode *nextop=(!ST.jump || !ST.jump[ST.accept_buff[best].wordnum]) ? + ST.B : + ST.me + ST.jump[ST.accept_buff[best].wordnum]; + PerlIO_printf( Perl_debug_log, + "%*s %strying alternation #%d <%s> at node #%d %s\n", + REPORT_CODE_OFF+depth*2, "", PL_colors[4], ST.accept_buff[best].wordnum, - tmp ? SvPV_nolen_const( *tmp ) : "not compiled under -Dr", REG_NODE_NUM(scan), + tmp ? SvPV_nolen_const( *tmp ) : "not compiled under -Dr", + REG_NODE_NUM(nextop), PL_colors[5] ); }); @@ -2735,19 +2958,26 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) best = ST.accepted; } PL_reginput = (char *)ST.accept_buff[ best ].endpos; + if ( !ST.jump || !ST.jump[ST.accept_buff[best].wordnum]) { + PUSH_STATE_GOTO(TRIE_next, ST.B); + /* NOTREACHED */ + } else { + PUSH_STATE_GOTO(TRIE_next, ST.me + ST.jump[ST.accept_buff[best].wordnum]); + /* NOTREACHED */ + } + /* NOTREACHED */ } - PUSH_STATE_GOTO(TRIE_next, ST.B); /* NOTREACHED */ #undef ST case EXACT: { char *s = STRING(scan); - st->ln = STR_LEN(scan); + ln = STR_LEN(scan); if (do_utf8 != UTF) { /* The target and the pattern have differing utf8ness. */ char *l = locinput; - const char * const e = s + st->ln; + const char * const e = s + ln; if (do_utf8) { /* The target is utf8, the pattern is not utf8. */ @@ -2785,11 +3015,11 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) /* Inline the first character, for speed. */ if (UCHARAT(s) != nextchr) sayNO; - if (PL_regeol - locinput < st->ln) + if (PL_regeol - locinput < ln) sayNO; - if (st->ln > 1 && memNE(s, locinput, st->ln)) + if (ln > 1 && memNE(s, locinput, ln)) sayNO; - locinput += st->ln; + locinput += ln; nextchr = UCHARAT(locinput); break; } @@ -2798,14 +3028,14 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) /* FALL THROUGH */ case EXACTF: { char * const s = STRING(scan); - st->ln = STR_LEN(scan); + ln = STR_LEN(scan); if (do_utf8 || UTF) { /* Either target or the pattern are utf8. */ const char * const l = locinput; char *e = PL_regeol; - if (ibcmp_utf8(s, 0, st->ln, (bool)UTF, + if (ibcmp_utf8(s, 0, ln, (bool)UTF, l, &e, 0, do_utf8)) { /* One more case for the sharp s: * pack("U0U*", 0xDF) =~ /ss/i, @@ -2813,7 +3043,7 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) * byte sequence for the U+00DF. */ if (!(do_utf8 && toLOWER(s[0]) == 's' && - st->ln >= 2 && + ln >= 2 && toLOWER(s[1]) == 's' && (U8)l[0] == 0xC3 && e - l >= 2 && @@ -2832,13 +3062,13 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) UCHARAT(s) != ((OP(scan) == EXACTF) ? PL_fold : PL_fold_locale)[nextchr]) sayNO; - if (PL_regeol - locinput < st->ln) + if (PL_regeol - locinput < ln) sayNO; - if (st->ln > 1 && (OP(scan) == EXACTF - ? ibcmp(s, locinput, st->ln) - : ibcmp_locale(s, locinput, st->ln))) + if (ln > 1 && (OP(scan) == EXACTF + ? ibcmp(s, locinput, ln) + : ibcmp_locale(s, locinput, ln))) sayNO; - locinput += st->ln; + locinput += ln; nextchr = UCHARAT(locinput); break; } @@ -2847,7 +3077,7 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) STRLEN inclasslen = PL_regeol - locinput; if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, do_utf8)) - sayNO_ANYOF; + goto anyof_fail; if (locinput >= PL_regeol) sayNO; locinput += inclasslen ? inclasslen : UTF8SKIP(locinput); @@ -2858,13 +3088,13 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) if (nextchr < 0) nextchr = UCHARAT(locinput); if (!REGINCLASS(rex, scan, (U8*)locinput)) - sayNO_ANYOF; + goto anyof_fail; if (!nextchr && locinput >= PL_regeol) sayNO; nextchr = UCHARAT(++locinput); break; } - no_anyof: + anyof_fail: /* If we might have the case of the German sharp s * in a casefolding Unicode character class. */ @@ -2930,35 +3160,35 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) /* was last char in word? */ if (do_utf8) { if (locinput == PL_bostr) - st->ln = '\n'; + ln = '\n'; else { const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr); - st->ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags); + ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags); } if (OP(scan) == BOUND || OP(scan) == NBOUND) { - st->ln = isALNUM_uni(st->ln); + ln = isALNUM_uni(ln); LOAD_UTF8_CHARCLASS_ALNUM(); n = swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8); } else { - st->ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(st->ln)); + ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln)); n = isALNUM_LC_utf8((U8*)locinput); } } else { - st->ln = (locinput != PL_bostr) ? + ln = (locinput != PL_bostr) ? UCHARAT(locinput - 1) : '\n'; if (OP(scan) == BOUND || OP(scan) == NBOUND) { - st->ln = isALNUM(st->ln); + ln = isALNUM(ln); n = isALNUM(nextchr); } else { - st->ln = isALNUM_LC(st->ln); + ln = isALNUM_LC(ln); n = isALNUM_LC(nextchr); } } - if (((!st->ln) == (!n)) == (OP(scan) == BOUND || + if (((!ln) == (!n)) == (OP(scan) == BOUND || OP(scan) == BOUNDL)) sayNO; break; @@ -3080,22 +3310,42 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) locinput++; nextchr = UCHARAT(locinput); break; + + case NREFFL: + { + char *s; + char type; + PL_reg_flags |= RF_tainted; + /* FALL THROUGH */ + case NREF: + case NREFF: + type = OP(scan); + n = reg_check_named_buff_matched(rex,scan); + + if ( n ) { + type = REF + ( type - NREF ); + goto do_ref; + } else { + sayNO; + } + /* unreached */ case REFFL: PL_reg_flags |= RF_tainted; /* FALL THROUGH */ case REF: - case REFF: { - char *s; + case REFF: n = ARG(scan); /* which paren pair */ - st->ln = PL_regstartp[n]; + type = OP(scan); + do_ref: + ln = PL_regstartp[n]; PL_reg_leftiter = PL_reg_maxiter; /* Void cache */ - if ((I32)*PL_reglastparen < n || st->ln == -1) + if ((I32)*PL_reglastparen < n || ln == -1) sayNO; /* Do not match unless seen CLOSEn. */ - if (st->ln == PL_regendp[n]) + if (ln == PL_regendp[n]) break; - s = PL_bostr + st->ln; - if (do_utf8 && OP(scan) != REF) { /* REF can do byte comparison */ + s = PL_bostr + ln; + if (do_utf8 && type != REF) { /* REF can do byte comparison */ char *l = locinput; const char *e = PL_bostr + PL_regendp[n]; /* @@ -3103,7 +3353,7 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) * in the 8-bit case (no pun intended) because in Unicode we * have to map both upper and title case to lower case. */ - if (OP(scan) == REFF) { + if (type == REFF) { while (s < e) { STRLEN ulen1, ulen2; U8 tmpbuf1[UTF8_MAXBYTES_CASE+1]; @@ -3126,24 +3376,23 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) /* Inline the first character, for speed. */ if (UCHARAT(s) != nextchr && - (OP(scan) == REF || - (UCHARAT(s) != ((OP(scan) == REFF - ? PL_fold : PL_fold_locale)[nextchr])))) + (type == REF || + (UCHARAT(s) != (type == REFF + ? PL_fold : PL_fold_locale)[nextchr]))) sayNO; - st->ln = PL_regendp[n] - st->ln; - if (locinput + st->ln > PL_regeol) + ln = PL_regendp[n] - ln; + if (locinput + ln > PL_regeol) sayNO; - if (st->ln > 1 && (OP(scan) == REF - ? memNE(s, locinput, st->ln) - : (OP(scan) == REFF - ? ibcmp(s, locinput, st->ln) - : ibcmp_locale(s, locinput, st->ln)))) + if (ln > 1 && (type == REF + ? memNE(s, locinput, ln) + : (type == REFF + ? ibcmp(s, locinput, ln) + : ibcmp_locale(s, locinput, ln)))) sayNO; - locinput += st->ln; + locinput += ln; nextchr = UCHARAT(locinput); break; - } - + } case NOTHING: case TAIL: break; @@ -3152,10 +3401,39 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) #undef ST #define ST st->u.eval - - case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */ { SV *ret; + regexp *re; + regnode *startpoint; + + case SRECURSE: + case RECURSE: /* /(...(?1))/ */ + if (cur_eval && cur_eval->locinput==locinput) { + if (cur_eval->u.eval.close_paren == ARG(scan)) + Perl_croak(aTHX_ "Infinite recursion in RECURSE in regexp"); + if ( ++nochange_depth > MAX_RECURSE_EVAL_NOCHANGE_DEPTH ) + Perl_croak(aTHX_ "RECURSE without pos change exceeded limit in regexp"); + } else { + nochange_depth = 0; + } + re = rex; + (void)ReREFCNT_inc(rex); + if (OP(scan)==RECURSE) { + startpoint = scan + ARG2L(scan); + ST.close_paren = ARG(scan); + } else { + startpoint = re->program+1; + ST.close_paren = 0; + } + goto eval_recurse_doit; + /* NOTREACHED */ + case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */ + if (cur_eval && cur_eval->locinput==locinput) { + if ( ++nochange_depth > MAX_RECURSE_EVAL_NOCHANGE_DEPTH ) + Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regexp"); + } else { + nochange_depth = 0; + } { /* execute the code in the {...} */ dSP; @@ -3182,14 +3460,14 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) PL_op = oop; PAD_RESTORE_LOCAL(old_comppad); PL_curcop = ocurcop; - if (!st->logical) { + if (!logical) { /* /(?{...})/ */ sv_setsv(save_scalar(PL_replgv), ret); break; } } - if (st->logical == 2) { /* Postponed subexpression: /(??{...})/ */ - regexp *re; + if (logical == 2) { /* Postponed subexpression: /(??{...})/ */ + logical = 0; { /* extract RE object from returned value; compiling if * necessary */ @@ -3217,7 +3495,7 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) Zero(&pm, 1, PMOP); if (DO_UTF8(ret)) pm.op_pmdynflags |= PMdf_DYN_UTF8; - re = CALLREGCOMP(aTHX_ (char*)t, (char*)t + len, &pm); + re = CALLREGCOMP((char*)t, (char*)t + len, &pm); if (!(SvFLAGS(ret) & (SVs_TEMP | SVs_PADTMP | SVf_READONLY | SVs_GMG))) @@ -3226,75 +3504,87 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) PL_regsize = osize; } } - - /* run the pattern returned from (??{...}) */ - DEBUG_EXECUTE_r( + DEBUG_EXECUTE_r( debug_start_match(re, do_utf8, locinput, PL_regeol, "Matching embedded"); - ); - + ); + startpoint = re->program + 1; + ST.close_paren = 0; /* only used for RECURSE */ + /* borrowed from regtry */ + if (PL_reg_start_tmpl <= re->nparens) { + PL_reg_start_tmpl = re->nparens*3/2 + 3; + if(PL_reg_start_tmp) + Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*); + else + Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*); + } + + eval_recurse_doit: /* Share code with RECURSE below this line */ + /* run the pattern returned from (??{...}) */ ST.cp = regcppush(0); /* Save *all* the positions. */ REGCP_SET(ST.lastcp); + + PL_regstartp = re->startp; /* essentially NOOP on RECURSE */ + PL_regendp = re->endp; /* essentially NOOP on RECURSE */ + *PL_reglastparen = 0; *PL_reglastcloseparen = 0; PL_reginput = locinput; + PL_regsize = 0; /* XXXX This is too dramatic a measure... */ PL_reg_maxiter = 0; - st->logical = 0; - ST.toggleutf = ((PL_reg_flags & RF_utf8) != 0) ^ - ((re->reganch & ROPT_UTF8) != 0); - if (ST.toggleutf) PL_reg_flags ^= RF_utf8; + ST.toggle_reg_flags = PL_reg_flags; + if (re->reganch & ROPT_UTF8) + PL_reg_flags |= RF_utf8; + else + PL_reg_flags &= ~RF_utf8; + ST.toggle_reg_flags ^= PL_reg_flags; /* diff of old and new */ + ST.prev_rex = rex; + ST.prev_curlyx = cur_curlyx; rex = re; - + cur_curlyx = NULL; ST.B = next; - /* now continue from first node in postoned RE */ - PUSH_YES_STATE_GOTO(EVAL_A, re->program + 1); + ST.prev_eval = cur_eval; + cur_eval = st; + /* now continue from first node in postoned RE */ + PUSH_YES_STATE_GOTO(EVAL_AB, startpoint); /* NOTREACHED */ } - /* /(?(?{...})X|Y)/ */ - st->sw = SvTRUE(ret); - st->logical = 0; + /* logical is 1, /(?(?{...})X|Y)/ */ + sw = (bool)SvTRUE(ret); + logical = 0; break; } - case EVAL_A: /* successfully ran inner rex (??{rex}) */ - if (ST.toggleutf) - PL_reg_flags ^= RF_utf8; + case EVAL_AB: /* cleanup after a successful (??{A})B */ + /* note: this is called twice; first after popping B, then A */ + PL_reg_flags ^= ST.toggle_reg_flags; ReREFCNT_dec(rex); rex = ST.prev_rex; + regcpblow(ST.cp); + cur_eval = ST.prev_eval; + cur_curlyx = ST.prev_curlyx; /* XXXX This is too dramatic a measure... */ PL_reg_maxiter = 0; - /* Restore parens of the caller without popping the - * savestack */ - { - const I32 tmp = PL_savestack_ix; - PL_savestack_ix = ST.lastcp; - regcppop(rex); - PL_savestack_ix = tmp; - } - PL_reginput = locinput; - /* continue at the node following the (??{...}) */ - scan = ST.B; - continue; + sayYES; + - case EVAL_A_fail: /* unsuccessfully ran inner rex (??{rex}) */ - /* Restore state to the outer re then re-throw the failure */ - if (ST.toggleutf) - PL_reg_flags ^= RF_utf8; + case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */ + /* note: this is called twice; first after popping B, then A */ + PL_reg_flags ^= ST.toggle_reg_flags; ReREFCNT_dec(rex); rex = ST.prev_rex; - - /* XXXX This is too dramatic a measure... */ - PL_reg_maxiter = 0; - PL_reginput = locinput; REGCP_UNWIND(ST.lastcp); regcppop(rex); + cur_eval = ST.prev_eval; + cur_curlyx = ST.prev_curlyx; + /* XXXX This is too dramatic a measure... */ + PL_reg_maxiter = 0; sayNO_SILENT; - #undef ST case OPEN: @@ -3307,17 +3597,33 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) n = ARG(scan); /* which paren pair */ PL_regstartp[n] = PL_reg_start_tmp[n] - PL_bostr; PL_regendp[n] = locinput - PL_bostr; + /*if (n > PL_regsize) + PL_regsize = n;*/ if (n > (I32)*PL_reglastparen) *PL_reglastparen = n; *PL_reglastcloseparen = n; + if (cur_eval && cur_eval->u.eval.close_paren == (U32)n) { + goto fake_end; + } break; case GROUPP: n = ARG(scan); /* which paren pair */ - st->sw = ((I32)*PL_reglastparen >= n && PL_regendp[n] != -1); + sw = (bool)((I32)*PL_reglastparen >= n && PL_regendp[n] != -1); + break; + case NGROUPP: + /* reg_check_named_buff_matched returns 0 for no match */ + sw = (bool)(0 < reg_check_named_buff_matched(rex,scan)); break; + case RECURSEP: + n = ARG(scan); + sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n)); + break; + case DEFINEP: + sw = 0; + break; case IFTHEN: PL_reg_leftiter = PL_reg_maxiter; /* Void cache */ - if (st->sw) + if (sw) next = NEXTOPER(NEXTOPER(scan)); else { next = scan + ARG(scan); @@ -3326,213 +3632,198 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) } break; case LOGICAL: - st->logical = scan->flags; + logical = scan->flags; break; + /******************************************************************* - cc points to the regmatch_state associated with the most recent CURLYX. - This struct contains info about the innermost (...)* loop (an - "infoblock"), and a pointer to the next outer cc. - - Here is how Y(A)*Z is processed (if it is compiled into CURLYX/WHILEM): - - 1) After matching Y, regnode for CURLYX is processed; - - 2) This regnode populates cc, and calls regmatch() recursively - with the starting point at WHILEM node; - - 3) Each hit of WHILEM node tries to match A and Z (in the order - depending on the current iteration, min/max of {min,max} and - greediness). The information about where are nodes for "A" - and "Z" is read from cc, as is info on how many times "A" - was already matched, and greediness. - - 4) After A matches, the same WHILEM node is hit again. - - 5) Each time WHILEM is hit, cc is the infoblock created by CURLYX - of the same pair. Thus when WHILEM tries to match Z, it temporarily - resets cc, since this Y(A)*Z can be a part of some other loop: - as in (Y(A)*Z)*. If Z matches, the automaton will hit the WHILEM node - of the external loop. - - Currently present infoblocks form a tree with a stem formed by st->cc - and whatever it mentions via ->next, and additional attached trees - corresponding to temporarily unset infoblocks as in "5" above. - - In the following picture, infoblocks for outer loop of - (Y(A)*?Z)*?T are denoted O, for inner I. NULL starting block - is denoted by x. The matched string is YAAZYAZT. Temporarily postponed - infoblocks are drawn below the "reset" infoblock. - - In fact in the picture below we do not show failed matches for Z and T - by WHILEM blocks. [We illustrate minimal matches, since for them it is - more obvious *why* one needs to *temporary* unset infoblocks.] - - Matched REx position InfoBlocks Comment - (Y(A)*?Z)*?T x - Y(A)*?Z)*?T x <- O - Y (A)*?Z)*?T x <- O - Y A)*?Z)*?T x <- O <- I - YA )*?Z)*?T x <- O <- I - YA A)*?Z)*?T x <- O <- I - YAA )*?Z)*?T x <- O <- I - YAA Z)*?T x <- O # Temporary unset I - I - - YAAZ Y(A)*?Z)*?T x <- O - I - - YAAZY (A)*?Z)*?T x <- O - I - - YAAZY A)*?Z)*?T x <- O <- I - I - - YAAZYA )*?Z)*?T x <- O <- I - I - - YAAZYA Z)*?T x <- O # Temporary unset I - I,I - - YAAZYAZ )*?T x <- O - I,I - - YAAZYAZ T x # Temporary unset O - O - I,I - - YAAZYAZT x - O - I,I - *******************************************************************/ - - case CURLYX: { - /* No need to save/restore up to this paren */ - I32 parenfloor = scan->flags; - - /* Dave says: - - CURLYX and WHILEM are always paired: they're the moral - equivalent of pp_enteriter anbd pp_iter. - - The only time next could be null is if the node tree is - corrupt. This was mentioned on p5p a few days ago. - - See http://www.xray.mpe.mpg.de/mailing-lists/perl5-porters/2006-04/msg00556.html - So we'll assert that this is true: - */ - assert(next); - if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */ - next += ARG(next); - /* XXXX Probably it is better to teach regpush to support - parenfloor > PL_regsize... */ - if (parenfloor > (I32)*PL_reglastparen) - parenfloor = *PL_reglastparen; /* Pessimization... */ - - st->u.curlyx.cp = PL_savestack_ix; - st->u.curlyx.outercc = st->cc; - st->cc = st; - /* these fields contain the state of the current curly. - * they are accessed by subsequent WHILEMs; - * cur and lastloc are also updated by WHILEM */ - st->u.curlyx.parenfloor = parenfloor; - st->u.curlyx.cur = -1; /* this will be updated by WHILEM */ - st->u.curlyx.min = ARG1(scan); - st->u.curlyx.max = ARG2(scan); - st->u.curlyx.scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS; - st->u.curlyx.lastloc = 0; - /* st->next and st->minmod are also read by WHILEM */ - PL_reginput = locinput; - REGMATCH(PREVOPER(next), CURLYX); /* start on the WHILEM */ - /*** all unsaved local vars undefined at this point */ - regcpblow(st->u.curlyx.cp); - st->cc = st->u.curlyx.outercc; - saySAME(result); - } +The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/ +pattern, where A and B are subpatterns. (For simple A, CURLYM or +STAR/PLUS/CURLY/CURLYN are used instead.) + +A*B is compiled as + +On entry to the subpattern, CURLYX is called. This pushes a CURLYX +state, which contains the current count, initialised to -1. It also sets +cur_curlyx to point to this state, with any previous value saved in the +state block. + +CURLYX then jumps straight to the WHILEM op, rather than executing A, +since the pattern may possibly match zero times (i.e. it's a while {} loop +rather than a do {} while loop). + +Each entry to WHILEM represents a successful match of A. The count in the +CURLYX block is incremented, another WHILEM state is pushed, and execution +passes to A or B depending on greediness and the current count. + +For example, if matching against the string a1a2a3b (where the aN are +substrings that match /A/), then the match progresses as follows: (the +pushed states are interspersed with the bits of strings matched so far): + + + + a1 + a1 a2 + a1 a2 a3 + a1 a2 a3 b + +(Contrast this with something like CURLYM, which maintains only a single +backtrack state: + + a1 + a1 a2 + a1 a2 a3 + a1 a2 a3 b +) + +Each WHILEM state block marks a point to backtrack to upon partial failure +of A or B, and also contains some minor state data related to that +iteration. The CURLYX block, pointed to by cur_curlyx, contains the +overall state, such as the count, and pointers to the A and B ops. + +This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx +must always point to the *current* CURLYX block, the rules are: + +When executing CURLYX, save the old cur_curlyx in the CURLYX state block, +and set cur_curlyx to point the new block. + +When popping the CURLYX block after a successful or unsuccessful match, +restore the previous cur_curlyx. + +When WHILEM is about to execute B, save the current cur_curlyx, and set it +to the outer one saved in the CURLYX block. + +When popping the WHILEM block after a successful or unsuccessful B match, +restore the previous cur_curlyx. + +Here's an example for the pattern (AI* BI)*BO +I and O refer to inner and outer, C and W refer to CURLYX and WHILEM: + +cur_ +curlyx backtrack stack +------ --------------- +NULL +CO +CI ai +CO ai bi +NULL ai bi bo + +At this point the pattern succeeds, and we work back down the stack to +clean up, restoring as we go: + +CO ai bi +CI ai +CO +NULL + +*******************************************************************/ + +#define ST st->u.curlyx + + case CURLYX: /* start of /A*B/ (for complex A) */ + { + /* No need to save/restore up to this paren */ + I32 parenfloor = scan->flags; + + assert(next); /* keep Coverity happy */ + if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */ + next += ARG(next); + + /* XXXX Probably it is better to teach regpush to support + parenfloor > PL_regsize... */ + if (parenfloor > (I32)*PL_reglastparen) + parenfloor = *PL_reglastparen; /* Pessimization... */ + + ST.prev_curlyx= cur_curlyx; + cur_curlyx = st; + ST.cp = PL_savestack_ix; + + /* these fields contain the state of the current curly. + * they are accessed by subsequent WHILEMs */ + ST.parenfloor = parenfloor; + ST.min = ARG1(scan); + ST.max = ARG2(scan); + ST.A = NEXTOPER(scan) + EXTRA_STEP_2ARGS; + ST.B = next; + ST.minmod = minmod; + minmod = 0; + ST.count = -1; /* this will be updated by WHILEM */ + ST.lastloc = NULL; /* this will be updated by WHILEM */ + + PL_reginput = locinput; + PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next)); /* NOTREACHED */ - case WHILEM: { - /* - * This is really hard to understand, because after we match - * what we're trying to match, we must make sure the rest of - * the REx is going to match for sure, and to do that we have - * to go back UP the parse tree by recursing ever deeper. And - * if it fails, we have to reset our parent's current state - * that we can try again after backing off. - */ + } - /* Dave says: - - st->cc gets initialised by CURLYX ready for use by WHILEM. - So again, unless somethings been corrupted, st->cc cannot - be null at that point in WHILEM. - - See http://www.xray.mpe.mpg.de/mailing-lists/perl5-porters/2006-04/msg00556.html - So we'll assert that this is true: - */ - assert(st->cc); - st->u.whilem.lastloc = st->cc->u.curlyx.lastloc; /* Detection of 0-len. */ - st->u.whilem.cache_offset = 0; - st->u.whilem.cache_bit = 0; - - n = st->cc->u.curlyx.cur + 1; /* how many we know we matched */ - PL_reginput = locinput; + case CURLYX_end: /* just finished matching all of A*B */ + regcpblow(ST.cp); + cur_curlyx = ST.prev_curlyx; + sayYES; + /* NOTREACHED */ - DEBUG_EXECUTE_r( - PerlIO_printf(Perl_debug_log, - "%*s %ld out of %ld..%ld cc=%"UVxf"\n", - REPORT_CODE_OFF+PL_regindent*2, "", - (long)n, (long)st->cc->u.curlyx.min, - (long)st->cc->u.curlyx.max, PTR2UV(st->cc)) - ); + case CURLYX_end_fail: /* just failed to match all of A*B */ + regcpblow(ST.cp); + cur_curlyx = ST.prev_curlyx; + sayNO; + /* NOTREACHED */ - /* If degenerate scan matches "", assume scan done. */ - - if (locinput == st->cc->u.curlyx.lastloc && n >= st->cc->u.curlyx.min) { - st->u.whilem.savecc = st->cc; - st->cc = st->cc->u.curlyx.outercc; - if (st->cc) - st->ln = st->cc->u.curlyx.cur; - DEBUG_EXECUTE_r( - PerlIO_printf(Perl_debug_log, - "%*s empty match detected, try continuation...\n", - REPORT_CODE_OFF+PL_regindent*2, "") - ); - REGMATCH(st->u.whilem.savecc->next, WHILEM1); - /*** all unsaved local vars undefined at this point */ - st->cc = st->u.whilem.savecc; - if (result) - sayYES; - if (st->cc->u.curlyx.outercc) - st->cc->u.curlyx.outercc->u.curlyx.cur = st->ln; - sayNO; - } - /* First just match a string of min scans. */ - - if (n < st->cc->u.curlyx.min) { - st->cc->u.curlyx.cur = n; - st->cc->u.curlyx.lastloc = locinput; - REGMATCH(st->cc->u.curlyx.scan, WHILEM2); - /*** all unsaved local vars undefined at this point */ - if (result) - sayYES; - st->cc->u.curlyx.cur = n - 1; - st->cc->u.curlyx.lastloc = st->u.whilem.lastloc; - sayNO; - } +#undef ST +#define ST st->u.whilem + + case WHILEM: /* just matched an A in /A*B/ (for complex A) */ + { + /* see the discussion above about CURLYX/WHILEM */ + I32 n; + assert(cur_curlyx); /* keep Coverity happy */ + n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */ + ST.save_lastloc = cur_curlyx->u.curlyx.lastloc; + ST.cache_offset = 0; + ST.cache_mask = 0; + + PL_reginput = locinput; + + DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, + "%*s whilem: matched %ld out of %ld..%ld\n", + REPORT_CODE_OFF+depth*2, "", (long)n, + (long)cur_curlyx->u.curlyx.min, + (long)cur_curlyx->u.curlyx.max) + ); + + /* First just match a string of min A's. */ + + if (n < cur_curlyx->u.curlyx.min) { + cur_curlyx->u.curlyx.lastloc = locinput; + PUSH_STATE_GOTO(WHILEM_A_pre, cur_curlyx->u.curlyx.A); + /* NOTREACHED */ + } + + /* If degenerate A matches "", assume A done. */ + + if (locinput == cur_curlyx->u.curlyx.lastloc) { + DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, + "%*s whilem: empty match detected, trying continuation...\n", + REPORT_CODE_OFF+depth*2, "") + ); + goto do_whilem_B_max; + } + + /* super-linear cache processing */ + + if (scan->flags) { - if (scan->flags) { - /* Check whether we already were at this position. - Postpone detection until we know the match is not - *that* much linear. */ if (!PL_reg_maxiter) { + /* start the countdown: Postpone detection until we + * know the match is not *that* much linear. */ PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4); /* possible overflow for long strings and many CURLYX's */ if (PL_reg_maxiter < 0) PL_reg_maxiter = I32_MAX; PL_reg_leftiter = PL_reg_maxiter; } + if (PL_reg_leftiter-- == 0) { + /* initialise cache */ const I32 size = (PL_reg_maxiter + 7)/8; if (PL_reg_poscache) { if ((I32)PL_reg_poscache_size < size) { @@ -3545,132 +3836,133 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) PL_reg_poscache_size = size; Newxz(PL_reg_poscache, size, char); } - DEBUG_EXECUTE_r( - PerlIO_printf(Perl_debug_log, - "%sDetected a super-linear match, switching on caching%s...\n", - PL_colors[4], PL_colors[5]) - ); + DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, + "%swhilem: Detected a super-linear match, switching on caching%s...\n", + PL_colors[4], PL_colors[5]) + ); } + if (PL_reg_leftiter < 0) { - st->u.whilem.cache_offset = locinput - PL_bostr; - - st->u.whilem.cache_offset = (scan->flags & 0xf) - 1 - + st->u.whilem.cache_offset * (scan->flags>>4); - st->u.whilem.cache_bit = st->u.whilem.cache_offset % 8; - st->u.whilem.cache_offset /= 8; - if (PL_reg_poscache[st->u.whilem.cache_offset] & (1<u.whilem.cache_bit)) { - DEBUG_EXECUTE_r( - PerlIO_printf(Perl_debug_log, - "%*s already tried at this position...\n", - REPORT_CODE_OFF+PL_regindent*2, "") + /* have we already failed at this position? */ + I32 offset, mask; + offset = (scan->flags & 0xf) - 1 + + (locinput - PL_bostr) * (scan->flags>>4); + mask = 1 << (offset % 8); + offset /= 8; + if (PL_reg_poscache[offset] & mask) { + DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, + "%*s whilem: (cache) already tried at this position...\n", + REPORT_CODE_OFF+depth*2, "") ); sayNO; /* cache records failure */ } + ST.cache_offset = offset; + ST.cache_mask = mask; } - } + } - /* Prefer next over scan for minimal matching. */ - - if (st->cc->minmod) { - st->u.whilem.savecc = st->cc; - st->cc = st->cc->u.curlyx.outercc; - if (st->cc) - st->ln = st->cc->u.curlyx.cur; - st->u.whilem.cp = regcppush(st->u.whilem.savecc->u.curlyx.parenfloor); - REGCP_SET(st->u.whilem.lastcp); - REGMATCH(st->u.whilem.savecc->next, WHILEM3); - /*** all unsaved local vars undefined at this point */ - st->cc = st->u.whilem.savecc; - if (result) { - regcpblow(st->u.whilem.cp); - sayYES; /* All done. */ - } - REGCP_UNWIND(st->u.whilem.lastcp); - regcppop(rex); - if (st->cc->u.curlyx.outercc) - st->cc->u.curlyx.outercc->u.curlyx.cur = st->ln; - - if (n >= st->cc->u.curlyx.max) { /* Maximum greed exceeded? */ - if (ckWARN(WARN_REGEXP) && n >= REG_INFTY - && !(PL_reg_flags & RF_warned)) { - PL_reg_flags |= RF_warned; - Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded", - "Complex regular subexpression recursion", - REG_INFTY - 1); - } - CACHEsayNO; - } + /* Prefer B over A for minimal matching. */ - DEBUG_EXECUTE_r( - PerlIO_printf(Perl_debug_log, - "%*s trying longer...\n", - REPORT_CODE_OFF+PL_regindent*2, "") - ); - /* Try scanning more and see if it helps. */ - PL_reginput = locinput; - st->cc->u.curlyx.cur = n; - st->cc->u.curlyx.lastloc = locinput; - st->u.whilem.cp = regcppush(st->cc->u.curlyx.parenfloor); - REGCP_SET(st->u.whilem.lastcp); - REGMATCH(st->cc->u.curlyx.scan, WHILEM4); - /*** all unsaved local vars undefined at this point */ - if (result) { - regcpblow(st->u.whilem.cp); - sayYES; - } - REGCP_UNWIND(st->u.whilem.lastcp); - regcppop(rex); - st->cc->u.curlyx.cur = n - 1; - st->cc->u.curlyx.lastloc = st->u.whilem.lastloc; - CACHEsayNO; - } + if (cur_curlyx->u.curlyx.minmod) { + ST.save_curlyx = cur_curlyx; + cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx; + ST.cp = regcppush(ST.save_curlyx->u.curlyx.parenfloor); + REGCP_SET(ST.lastcp); + PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B); + /* NOTREACHED */ + } - /* Prefer scan over next for maximal matching. */ - - if (n < st->cc->u.curlyx.max) { /* More greed allowed? */ - st->u.whilem.cp = regcppush(st->cc->u.curlyx.parenfloor); - st->cc->u.curlyx.cur = n; - st->cc->u.curlyx.lastloc = locinput; - REGCP_SET(st->u.whilem.lastcp); - REGMATCH(st->cc->u.curlyx.scan, WHILEM5); - /*** all unsaved local vars undefined at this point */ - if (result) { - regcpblow(st->u.whilem.cp); - sayYES; - } - REGCP_UNWIND(st->u.whilem.lastcp); - regcppop(rex); /* Restore some previous $s? */ - PL_reginput = locinput; - DEBUG_EXECUTE_r( - PerlIO_printf(Perl_debug_log, - "%*s failed, try continuation...\n", - REPORT_CODE_OFF+PL_regindent*2, "") - ); - } - if (ckWARN(WARN_REGEXP) && n >= REG_INFTY - && !(PL_reg_flags & RF_warned)) { + /* Prefer A over B for maximal matching. */ + + if (n < cur_curlyx->u.curlyx.max) { /* More greed allowed? */ + ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor); + cur_curlyx->u.curlyx.lastloc = locinput; + REGCP_SET(ST.lastcp); + PUSH_STATE_GOTO(WHILEM_A_max, cur_curlyx->u.curlyx.A); + /* NOTREACHED */ + } + goto do_whilem_B_max; + } + /* NOTREACHED */ + + case WHILEM_B_min: /* just matched B in a minimal match */ + case WHILEM_B_max: /* just matched B in a maximal match */ + cur_curlyx = ST.save_curlyx; + sayYES; + /* NOTREACHED */ + + case WHILEM_B_max_fail: /* just failed to match B in a maximal match */ + cur_curlyx = ST.save_curlyx; + cur_curlyx->u.curlyx.lastloc = ST.save_lastloc; + cur_curlyx->u.curlyx.count--; + CACHEsayNO; + /* NOTREACHED */ + + case WHILEM_A_min_fail: /* just failed to match A in a minimal match */ + REGCP_UNWIND(ST.lastcp); + regcppop(rex); + /* FALL THROUGH */ + case WHILEM_A_pre_fail: /* just failed to match even minimal A */ + cur_curlyx->u.curlyx.lastloc = ST.save_lastloc; + cur_curlyx->u.curlyx.count--; + CACHEsayNO; + /* NOTREACHED */ + + case WHILEM_A_max_fail: /* just failed to match A in a maximal match */ + REGCP_UNWIND(ST.lastcp); + regcppop(rex); /* Restore some previous $s? */ + PL_reginput = locinput; + DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, + "%*s whilem: failed, trying continuation...\n", + REPORT_CODE_OFF+depth*2, "") + ); + do_whilem_B_max: + if (cur_curlyx->u.curlyx.count >= REG_INFTY + && ckWARN(WARN_REGEXP) + && !(PL_reg_flags & RF_warned)) + { + PL_reg_flags |= RF_warned; + Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded", + "Complex regular subexpression recursion", + REG_INFTY - 1); + } + + /* now try B */ + ST.save_curlyx = cur_curlyx; + cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx; + PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B); + /* NOTREACHED */ + + case WHILEM_B_min_fail: /* just failed to match B in a minimal match */ + cur_curlyx = ST.save_curlyx; + REGCP_UNWIND(ST.lastcp); + regcppop(rex); + + if (cur_curlyx->u.curlyx.count >= cur_curlyx->u.curlyx.max) { + /* Maximum greed exceeded */ + if (cur_curlyx->u.curlyx.count >= REG_INFTY + && ckWARN(WARN_REGEXP) + && !(PL_reg_flags & RF_warned)) + { PL_reg_flags |= RF_warned; - Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded", - "Complex regular subexpression recursion", - REG_INFTY - 1); + Perl_warner(aTHX_ packWARN(WARN_REGEXP), + "%s limit (%d) exceeded", + "Complex regular subexpression recursion", + REG_INFTY - 1); } - - /* Failed deeper matches of scan, so see if this one works. */ - st->u.whilem.savecc = st->cc; - st->cc = st->cc->u.curlyx.outercc; - if (st->cc) - st->ln = st->cc->u.curlyx.cur; - REGMATCH(st->u.whilem.savecc->next, WHILEM6); - /*** all unsaved local vars undefined at this point */ - st->cc = st->u.whilem.savecc; - if (result) - sayYES; - if (st->cc->u.curlyx.outercc) - st->cc->u.curlyx.outercc->u.curlyx.cur = st->ln; - st->cc->u.curlyx.cur = n - 1; - st->cc->u.curlyx.lastloc = st->u.whilem.lastloc; + cur_curlyx->u.curlyx.count--; CACHEsayNO; } + + DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, + "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "") + ); + /* Try grabbing another A and see if it helps. */ + PL_reginput = locinput; + cur_curlyx->u.curlyx.lastloc = locinput; + ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor); + REGCP_SET(ST.lastcp); + PUSH_STATE_GOTO(WHILEM_A_min, ST.save_curlyx->u.curlyx.A); /* NOTREACHED */ #undef ST @@ -3702,6 +3994,7 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) for (n = *PL_reglastparen; n > ST.lastparen; n--) PL_regendp[n] = -1; *PL_reglastparen = n; + /*dmq: *PL_reglastcloseparen = n; */ scan = ST.next_branch; /* no more branches? */ if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) @@ -3710,7 +4003,7 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) /* NOTREACHED */ case MINMOD: - st->minmod = 1; + minmod = 1; break; #undef ST @@ -3740,8 +4033,8 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) ST.B = next; ST.alen = 0; ST.count = 0; - ST.minmod = st->minmod; - st->minmod = 0; + ST.minmod = minmod; + minmod = 0; ST.c1 = CHRTEST_UNINIT; REGCP_SET(ST.cp); @@ -3776,18 +4069,26 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n", - (int)(REPORT_CODE_OFF+(PL_regindent*2)), "", + (int)(REPORT_CODE_OFF+(depth*2)), "", (IV) ST.count, (IV)ST.alen) ); locinput = PL_reginput; - if (ST.count < (ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) + + if (cur_eval && cur_eval->u.eval.close_paren && + cur_eval->u.eval.close_paren == ST.me->flags) + goto fake_end; + + if ( ST.count < (ST.minmod ? ARG1(ST.me) : ARG2(ST.me)) ) goto curlym_do_A; /* try to match another A */ goto curlym_do_B; /* try to match B */ case CURLYM_A_fail: /* just failed to match an A */ REGCP_UNWIND(ST.cp); - if (ST.minmod || ST.count < ARG1(ST.me) /* min*/ ) + + if (ST.minmod || ST.count < ARG1(ST.me) /* min*/ + || (cur_eval && cur_eval->u.eval.close_paren && + cur_eval->u.eval.close_paren == ST.me->flags)) sayNO; curlym_do_B: /* execute the B in /A{m,n}B/ */ @@ -3817,7 +4118,7 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, "%*s CURLYM trying tail with matches=%"IVdf"...\n", - (int)(REPORT_CODE_OFF+(PL_regindent*2)), + (int)(REPORT_CODE_OFF+(depth*2)), "", (IV)ST.count) ); if (ST.c1 != CHRTEST_VOID @@ -3836,10 +4137,20 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) PL_regstartp[paren] = HOPc(PL_reginput, -ST.alen) - PL_bostr; PL_regendp[paren] = PL_reginput - PL_bostr; + /*dmq: *PL_reglastcloseparen = paren; */ } else PL_regendp[paren] = -1; + if (cur_eval && cur_eval->u.eval.close_paren && + cur_eval->u.eval.close_paren == ST.me->flags) + { + if (ST.count) + goto fake_end; + else + sayNO; + } } + PUSH_STATE_GOTO(CURLYM_B, ST.B); /* match B */ /* NOTREACHED */ @@ -3865,6 +4176,7 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) if (success) { \ PL_regstartp[paren] = HOPc(locinput, -1) - PL_bostr; \ PL_regendp[paren] = locinput - PL_bostr; \ + *PL_reglastcloseparen = paren; \ } \ else \ PL_regendp[paren] = -1; \ @@ -3890,6 +4202,11 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) *PL_reglastparen = ST.paren; ST.min = ARG1(scan); /* min to match */ ST.max = ARG2(scan); /* max to match */ + if (cur_eval && cur_eval->u.eval.close_paren && + cur_eval->u.eval.close_paren == ST.paren) { + ST.min=1; + ST.max=1; + } scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE); goto repeat; case CURLY: /* /A{m,n}B/ where A is width 1 */ @@ -3941,11 +4258,19 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) to_utf8_lower((U8*)s, tmpbuf1, &ulen1); to_utf8_upper((U8*)s, tmpbuf2, &ulen2); - +#ifdef EBCDIC + ST.c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXLEN, 0, + ckWARN(WARN_UTF8) ? + 0 : UTF8_ALLOW_ANY); + ST.c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXLEN, 0, + ckWARN(WARN_UTF8) ? + 0 : UTF8_ALLOW_ANY); +#else ST.c1 = utf8n_to_uvuni(tmpbuf1, UTF8_MAXBYTES, 0, - uniflags); + uniflags); ST.c2 = utf8n_to_uvuni(tmpbuf2, UTF8_MAXBYTES, 0, - uniflags); + uniflags); +#endif } else { ST.c2 = ST.c1 = utf8n_to_uvchr(s, UTF8_MAXBYTES, 0, @@ -3961,8 +4286,8 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) ST.A = scan; ST.B = next; PL_reginput = locinput; - if (st->minmod) { - st->minmod = 0; + if (minmod) { + minmod = 0; if (ST.min && regrepeat(rex, ST.A, ST.min) < ST.min) sayNO; ST.count = ST.min; @@ -4087,6 +4412,10 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) } PL_reginput = locinput; CURLY_SETPAREN(ST.paren, ST.count); + if (cur_eval && cur_eval->u.eval.close_paren && + cur_eval->u.eval.close_paren == ST.paren) { + goto fake_end; + } PUSH_STATE_GOTO(CURLY_B_min_known, ST.B); } /* NOTREACHED */ @@ -4108,6 +4437,10 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) { curly_try_B_min: CURLY_SETPAREN(ST.paren, ST.count); + if (cur_eval && cur_eval->u.eval.close_paren && + cur_eval->u.eval.close_paren == ST.paren) { + goto fake_end; + } PUSH_STATE_GOTO(CURLY_B_min, ST.B); } } @@ -4126,6 +4459,10 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) /* If it could work, try it. */ if (ST.c1 == CHRTEST_VOID || c == (UV)ST.c1 || c == (UV)ST.c2) { CURLY_SETPAREN(ST.paren, ST.count); + if (cur_eval && cur_eval->u.eval.close_paren && + cur_eval->u.eval.close_paren == ST.paren) { + goto fake_end; + } PUSH_STATE_GOTO(CURLY_B_max, ST.B); /* NOTREACHED */ } @@ -4145,8 +4482,41 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) #undef ST - case END: + fake_end: + if (cur_eval) { + /* we've just finished A in /(??{A})B/; now continue with B */ + I32 tmpix; + + + st->u.eval.toggle_reg_flags + = cur_eval->u.eval.toggle_reg_flags; + PL_reg_flags ^= st->u.eval.toggle_reg_flags; + + st->u.eval.prev_rex = rex; /* inner */ + rex = cur_eval->u.eval.prev_rex; /* outer */ + cur_curlyx = cur_eval->u.eval.prev_curlyx; + ReREFCNT_inc(rex); + st->u.eval.cp = regcppush(0); /* Save *all* the positions. */ + REGCP_SET(st->u.eval.lastcp); + PL_reginput = locinput; + + /* Restore parens of the outer rex without popping the + * savestack */ + tmpix = PL_savestack_ix; + PL_savestack_ix = cur_eval->u.eval.lastcp; + regcppop(rex); + PL_savestack_ix = tmpix; + + st->u.eval.prev_eval = cur_eval; + cur_eval = cur_eval->u.eval.prev_eval; + DEBUG_EXECUTE_r( + PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %x\n", + REPORT_CODE_OFF+depth*2, "",(int)cur_eval);); + PUSH_YES_STATE_GOTO(EVAL_AB, + st->u.eval.prev_eval->u.eval.B); /* match B */ + } + if (locinput < reginfo->till) { DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n", @@ -4154,18 +4524,18 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) (long)(locinput - PL_reg_starttry), (long)(reginfo->till - PL_reg_starttry), PL_colors[5])); - sayNO_FINAL; /* Cannot match: too short. */ + sayNO_SILENT; /* Cannot match: too short. */ } PL_reginput = locinput; /* put where regtry can find it */ - sayYES_FINAL; /* Success! */ + sayYES; /* Success! */ case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */ DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, "%*s %ssubpattern success...%s\n", - REPORT_CODE_OFF+PL_regindent*2, "", PL_colors[4], PL_colors[5])); + REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])); PL_reginput = locinput; /* put where regtry can find it */ - sayYES_FINAL; /* Success! */ + sayYES; /* Success! */ #undef ST #define ST st->u.ifmatch @@ -4186,9 +4556,9 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) char * const s = HOPBACKc(locinput, scan->flags); if (!s) { /* trivial fail */ - if (st->logical) { - st->logical = 0; - st->sw = 1 - (bool)ST.wanted; + if (logical) { + logical = 0; + sw = 1 - (bool)ST.wanted; } else if (ST.wanted) sayNO; @@ -4204,6 +4574,7 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) do_ifmatch: ST.me = scan; + ST.logical = logical; /* execute body of (?...A) */ PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan))); /* NOTREACHED */ @@ -4213,9 +4584,8 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) /* FALL THROUGH */ case IFMATCH_A: /* body of (?...A) succeeded */ - if (st->logical) { - st->logical = 0; - st->sw = (bool)ST.wanted; + if (ST.logical) { + sw = (bool)ST.wanted; } else if (!ST.wanted) sayNO; @@ -4238,6 +4608,8 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) if (next == scan) next = NULL; break; + case OPFAIL: + sayNO; default: PerlIO_printf(Perl_error_log, "%"UVxf" %d\n", PTR2UV(scan), OP(scan)); @@ -4258,19 +4630,13 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) { regmatch_state *newst; + DEBUG_STATE_pp("push"); depth++; - DEBUG_STATE_r(PerlIO_printf(Perl_debug_log, - "PUSH STATE(%d)\n", depth)); st->locinput = locinput; newst = st+1; if (newst > SLAB_LAST(PL_regmatch_slab)) newst = S_push_slab(aTHX); PL_regmatch_state = newst; - newst->cc = st->cc; - /* XXX probably don't need to initialise these */ - newst->minmod = 0; - newst->sw = 0; - newst->logical = 0; locinput = PL_reginput; nextchr = UCHARAT(locinput); @@ -4278,44 +4644,8 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) continue; /* NOTREACHED */ } - - /* simulate recursively calling regmatch(), but without actually - * recursing - ie save the current state on the heap rather than on - * the stack, then re-enter the loop. This avoids complex regexes - * blowing the processor stack */ - - start_recurse: - { - /* push new state */ - regmatch_state *oldst = st; - - depth++; - DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "PUSH RECURSE STATE(%d)\n", depth)); - - /* grab the next free state slot */ - st++; - if (st > SLAB_LAST(PL_regmatch_slab)) - st = S_push_slab(aTHX); - PL_regmatch_state = st; - - oldst->next = next; - oldst->n = n; - oldst->locinput = locinput; - - st->cc = oldst->cc; - locinput = PL_reginput; - nextchr = UCHARAT(locinput); - st->minmod = 0; - st->sw = 0; - st->logical = 0; -#ifdef DEBUGGING - PL_regindent++; -#endif - } } - - /* * We get here only if there's trouble -- normally "case END" is * the terminating point. @@ -4324,13 +4654,22 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) /*NOTREACHED*/ sayNO; -yes_final: - +yes: if (yes_state) { /* we have successfully completed a subexpression, but we must now * pop to the state marked by yes_state and continue from there */ - assert(st != yes_state); +#ifdef DEBUGGING + while (st != yes_state) { + st--; + if (st < SLAB_FIRST(PL_regmatch_slab)) { + PL_regmatch_slab = PL_regmatch_slab->prev; + st = SLAB_LAST(PL_regmatch_slab); + } + DEBUG_STATE_pp("pop (yes)"); + depth--; + } +#else while (yes_state < SLAB_FIRST(PL_regmatch_slab) || yes_state > SLAB_LAST(PL_regmatch_slab)) { @@ -4340,153 +4679,52 @@ yes_final: st = SLAB_LAST(PL_regmatch_slab); } depth -= (st - yes_state); - DEBUG_STATE_r(PerlIO_printf(Perl_debug_log, "POP STATES (%d..%d)\n", - depth+1, depth+(st - yes_state))); +#endif st = yes_state; yes_state = st->u.yes.prev_yes_state; PL_regmatch_state = st; - switch (st->resume_state) { - case IFMATCH_A: - case CURLYM_A: - case EVAL_A: - state_num = st->resume_state; - goto reenter_switch; - - case CURLYM_B: - case BRANCH_next: - case TRIE_next: - case CURLY_B_max: - default: - Perl_croak(aTHX_ "unexpected yes resume state"); - } + state_num = st->resume_state; + goto reenter_switch; } DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n", PL_colors[4], PL_colors[5])); -yes: -#ifdef DEBUGGING - PL_regindent--; -#endif result = 1; - /* XXX this is duplicate(ish) code to that in the do_no section. - * will disappear when REGFMATCH goes */ - if (depth) { - /* restore previous state and re-enter */ - DEBUG_STATE_r(PerlIO_printf(Perl_debug_log, "POP STATE(%d)\n", depth)); - depth--; - st--; - if (st < SLAB_FIRST(PL_regmatch_slab)) { - PL_regmatch_slab = PL_regmatch_slab->prev; - st = SLAB_LAST(PL_regmatch_slab); - } - PL_regmatch_state = st; - scan = st->scan; - next = st->next; - n = st->n; - locinput= st->locinput; - nextchr = UCHARAT(locinput); - - switch (st->resume_state) { - case resume_CURLYX: - goto resume_point_CURLYX; - case resume_WHILEM1: - goto resume_point_WHILEM1; - case resume_WHILEM2: - goto resume_point_WHILEM2; - case resume_WHILEM3: - goto resume_point_WHILEM3; - case resume_WHILEM4: - goto resume_point_WHILEM4; - case resume_WHILEM5: - goto resume_point_WHILEM5; - case resume_WHILEM6: - goto resume_point_WHILEM6; - - case TRIE_next: - case CURLYM_A: - case CURLYM_B: - case EVAL_A: - case IFMATCH_A: - case BRANCH_next: - case CURLY_B_max: - case CURLY_B_min: - case CURLY_B_min_known: - break; - - default: - Perl_croak(aTHX_ "regexp resume memory corruption"); - } - } goto final_exit; no: DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, - "%*s %sfailed...%s\n", - REPORT_CODE_OFF+PL_regindent*2, "", PL_colors[4], PL_colors[5]) + "%*s %sfailed...%s\n", + REPORT_CODE_OFF+depth*2, "", + PL_colors[4], PL_colors[5]) ); -no_final: -do_no: - -#ifdef DEBUGGING - PL_regindent--; -#endif - result = 0; +no_silent: if (depth) { /* there's a previous state to backtrack to */ - DEBUG_STATE_r(PerlIO_printf(Perl_debug_log, "POP STATE(%d)\n", depth)); - depth--; st--; if (st < SLAB_FIRST(PL_regmatch_slab)) { PL_regmatch_slab = PL_regmatch_slab->prev; st = SLAB_LAST(PL_regmatch_slab); } PL_regmatch_state = st; - scan = st->scan; - next = st->next; - n = st->n; locinput= st->locinput; nextchr = UCHARAT(locinput); - switch (st->resume_state) { - case resume_CURLYX: - goto resume_point_CURLYX; - case resume_WHILEM1: - goto resume_point_WHILEM1; - case resume_WHILEM2: - goto resume_point_WHILEM2; - case resume_WHILEM3: - goto resume_point_WHILEM3; - case resume_WHILEM4: - goto resume_point_WHILEM4; - case resume_WHILEM5: - goto resume_point_WHILEM5; - case resume_WHILEM6: - goto resume_point_WHILEM6; - - case TRIE_next: - case EVAL_A: - case BRANCH_next: - case CURLYM_A: - case CURLYM_B: - case IFMATCH_A: - case CURLY_B_max: - case CURLY_B_min: - case CURLY_B_min_known: - if (yes_state == st) - yes_state = st->u.yes.prev_yes_state; - state_num = st->resume_state + 1; /* failure = success + 1 */ - goto reenter_switch; + DEBUG_STATE_pp("pop"); + depth--; + if (yes_state == st) + yes_state = st->u.yes.prev_yes_state; - default: - Perl_croak(aTHX_ "regexp resume memory corruption"); - } + state_num = st->resume_state + 1; /* failure = success + 1 */ + goto reenter_switch; } + result = 0; -final_exit: + final_exit: /* restore original high-water mark */ PL_regmatch_slab = orig_slab; @@ -4504,7 +4742,6 @@ final_exit: } return result; - } /* @@ -4829,7 +5066,7 @@ S_reginclass(pTHX_ const regexp *prog, register const regnode *n, register const c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &len, (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV) | UTF8_CHECK_ONLY); /* see [perl #37836] for UTF8_ALLOW_ANYUV */ - if (len == (STRLEN)-1) + if (len == (STRLEN)-1) Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)"); } @@ -4949,20 +5186,47 @@ S_reghop3(U8 *s, I32 off, const U8* lim) } } else { - while (off++) { - if (s > lim) { - s--; - if (UTF8_IS_CONTINUED(*s)) { - while (s > (U8*)lim && UTF8_IS_CONTINUATION(*s)) - s--; - } - /* XXX could check well-formedness here */ + while (off++ && s > lim) { + s--; + if (UTF8_IS_CONTINUED(*s)) { + while (s > lim && UTF8_IS_CONTINUATION(*s)) + s--; } + /* XXX could check well-formedness here */ } } return s; } +#ifdef XXX_dmq +/* there are a bunch of places where we use two reghop3's that should + be replaced with this routine. but since thats not done yet + we ifdef it out - dmq +*/ +STATIC U8 * +S_reghop4(U8 *s, I32 off, const U8* llim, const U8* rlim) +{ + dVAR; + if (off >= 0) { + while (off-- && s < rlim) { + /* XXX could check well-formedness here */ + s += UTF8SKIP(s); + } + } + else { + while (off++ && s > llim) { + s--; + if (UTF8_IS_CONTINUED(*s)) { + while (s > llim && UTF8_IS_CONTINUATION(*s)) + s--; + } + /* XXX could check well-formedness here */ + } + } + return s; +} +#endif + STATIC U8 * S_reghopmaybe3(U8* s, I32 off, const U8* lim) { @@ -4976,17 +5240,13 @@ S_reghopmaybe3(U8* s, I32 off, const U8* lim) return NULL; } else { - while (off++) { - if (s > lim) { - s--; - if (UTF8_IS_CONTINUED(*s)) { - while (s > (U8*)lim && UTF8_IS_CONTINUATION(*s)) - s--; - } - /* XXX could check well-formedness here */ + while (off++ && s > lim) { + s--; + if (UTF8_IS_CONTINUED(*s)) { + while (s > lim && UTF8_IS_CONTINUATION(*s)) + s--; } - else - break; + /* XXX could check well-formedness here */ } if (off <= 0) return NULL;